langchain 0.0.203 → 0.0.204-rc.1

package/dist/chat_models/llama_cpp.cjs

@@ -1,277 +1,17 @@
 "use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __exportStar = (this && this.__exportStar) || function(m, exports) {
+    for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
+};
 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");
-const llama_cpp_js_1 = require("../util/llama_cpp.cjs");
-const index_js_1 = require("../schema/index.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.
- * @example
- * ```typescript
- * // Initialize the ChatLlamaCpp model with the path to the model binary file.
- * const model = new ChatLlamaCpp({
- *   modelPath: "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin",
- *   temperature: 0.5,
- * });
- *
- * // Call the model with a message and await the response.
- * const response = await model.call([
- *   new HumanMessage({ content: "My name is John." }),
- * ]);
- *
- * // Log the response to the console.
- * console.log({ response });
- *
- * ```
- */
-class ChatLlamaCpp extends base_js_1.SimpleChatModel {
-    static lc_name() {
-        return "ChatLlamaCpp";
-    }
-    constructor(inputs) {
-        super(inputs);
-        Object.defineProperty(this, "maxTokens", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "temperature", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "topK", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "topP", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "trimWhitespaceSuffix", {
-            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.maxTokens = inputs?.maxTokens;
-        this.temperature = inputs?.temperature;
-        this.topK = inputs?.topK;
-        this.topP = inputs?.topP;
-        this.trimWhitespaceSuffix = inputs?.trimWhitespaceSuffix;
-        this._model = (0, llama_cpp_js_1.createLlamaModel)(inputs);
-        this._context = (0, llama_cpp_js_1.createLlamaContext)(this._model, inputs);
-        this._session = null;
-    }
-    _llmType() {
-        return "llama2_cpp";
-    }
-    /** @ignore */
-    _combineLLMOutput() {
-        return {};
-    }
-    invocationParams() {
-        return {
-            maxTokens: this.maxTokens,
-            temperature: this.temperature,
-            topK: this.topK,
-            topP: this.topP,
-            trimWhitespaceSuffix: this.trimWhitespaceSuffix,
-        };
-    }
-    /** @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 (typeof messages[0].content !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // If we already have a session then we should just have a single prompt
-            prompt = messages[0].content;
-        }
-        try {
-            const promptOptions = {
-                onToken: options.onToken,
-                maxTokens: this?.maxTokens,
-                temperature: this?.temperature,
-                topK: this?.topK,
-                topP: this?.topP,
-                trimWhitespaceSuffix: this?.trimWhitespaceSuffix,
-            };
-            // @ts-expect-error - TS2531: Object is possibly 'null'.
-            const completion = await this._session.prompt(prompt, promptOptions);
-            return completion;
-        }
-        catch (e) {
-            throw new Error("Error getting prompt completion.");
-        }
-    }
-    async *_streamResponseChunks(input, _options, runManager) {
-        const promptOptions = {
-            temperature: this?.temperature,
-            topK: this?.topK,
-            topP: this?.topP,
-        };
-        const prompt = this._buildPrompt(input);
-        const stream = await this.caller.call(async () => this._context.evaluate(this._context.encode(prompt), promptOptions));
-        for await (const chunk of stream) {
-            yield new index_js_1.ChatGenerationChunk({
-                text: this._context.decode([chunk]),
-                message: new index_js_1.AIMessageChunk({
-                    content: this._context.decode([chunk]),
-                }),
-                generationInfo: {},
-            });
-            await runManager?.handleLLMNewToken(this._context.decode([chunk]) ?? "");
-        }
-    }
-    // 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");
-            const systemMessageContent = sysMessages[sysMessages.length - 1].content;
-            if (typeof systemMessageContent !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // Only use the last provided system message
-            sysMessage = systemMessageContent;
-            // 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") {
-                const finalMessageContent = noSystemMessages[noSystemMessages.length - 1].content;
-                if (typeof finalMessageContent !== "string") {
-                    throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-                }
-                prompt = finalMessageContent;
-                interactions = this._convertMessagesToInteractions(noSystemMessages.slice(0, noSystemMessages.length - 1));
-            }
-            else {
-                interactions = this._convertMessagesToInteractions(noSystemMessages);
-            }
-        }
-        else {
-            if (typeof noSystemMessages[0].content !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // 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) {
-                const prompt = messages[i].content;
-                const response = messages[i + 1].content;
-                if (typeof prompt !== "string" || typeof response !== "string") {
-                    throw new Error("ChatLlamaCpp does not support non-string message content.");
-                }
-                result.push({
-                    prompt,
-                    response,
-                });
-            }
-        }
-        return result;
-    }
-    _buildPrompt(input) {
-        const prompt = input
-            .map((message) => {
-            let messageText;
-            if (message._getType() === "human") {
-                messageText = `[INST] ${message.content} [/INST]`;
-            }
-            else if (message._getType() === "ai") {
-                messageText = message.content;
-            }
-            else if (message._getType() === "system") {
-                messageText = `<<SYS>> ${message.content} <</SYS>>`;
-            }
-            else if (index_js_1.ChatMessage.isInstance(message)) {
-                messageText = `\n\n${message.role[0].toUpperCase()}${message.role.slice(1)}: ${message.content}`;
-            }
-            else {
-                console.warn(`Unsupported message type passed to llama_cpp: "${message._getType()}"`);
-                messageText = "";
-            }
-            return messageText;
-        })
-            .join("\n");
-        return prompt;
-    }
-}
-exports.ChatLlamaCpp = ChatLlamaCpp;
+__exportStar(require("@langchain/community/chat_models/llama_cpp"), exports);

package/dist/chat_models/llama_cpp.d.ts

@@ -1,71 +1 @@
-import { LlamaModel, LlamaContext, LlamaChatSession, type ConversationInteraction } from "node-llama-cpp";
-import { SimpleChatModel, BaseChatModelParams } from "./base.js";
-import { LlamaBaseCppInputs } from "../util/llama_cpp.js";
-import { BaseLanguageModelCallOptions } from "../base_language/index.js";
-import { CallbackManagerForLLMRun } from "../callbacks/manager.js";
-import { BaseMessage, ChatGenerationChunk } 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 LlamaBaseCppInputs, BaseChatModelParams {
-}
-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.
- * @example
- * ```typescript
- * // Initialize the ChatLlamaCpp model with the path to the model binary file.
- * const model = new ChatLlamaCpp({
- *   modelPath: "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin",
- *   temperature: 0.5,
- * });
- *
- * // Call the model with a message and await the response.
- * const response = await model.call([
- *   new HumanMessage({ content: "My name is John." }),
- * ]);
- *
- * // Log the response to the console.
- * console.log({ response });
- *
- * ```
- */
-export declare class ChatLlamaCpp extends SimpleChatModel<LlamaCppCallOptions> {
-    CallOptions: LlamaCppCallOptions;
-    static inputs: LlamaCppInputs;
-    maxTokens?: number;
-    temperature?: number;
-    topK?: number;
-    topP?: number;
-    trimWhitespaceSuffix?: boolean;
-    _model: LlamaModel;
-    _context: LlamaContext;
-    _session: LlamaChatSession | null;
-    static lc_name(): string;
-    constructor(inputs: LlamaCppInputs);
-    _llmType(): string;
-    /** @ignore */
-    _combineLLMOutput(): {};
-    invocationParams(): {
-        maxTokens: number | undefined;
-        temperature: number | undefined;
-        topK: number | undefined;
-        topP: number | undefined;
-        trimWhitespaceSuffix: boolean | undefined;
-    };
-    /** @ignore */
-    _call(messages: BaseMessage[], options: this["ParsedCallOptions"]): Promise<string>;
-    _streamResponseChunks(input: BaseMessage[], _options: this["ParsedCallOptions"], runManager?: CallbackManagerForLLMRun): AsyncGenerator<ChatGenerationChunk>;
-    protected _buildSession(messages: BaseMessage[]): string;
-    protected _convertMessagesToInteractions(messages: BaseMessage[]): ConversationInteraction[];
-    protected _buildPrompt(input: BaseMessage[]): string;
-}
+export * from "@langchain/community/chat_models/llama_cpp";

package/dist/chat_models/llama_cpp.js

@@ -1,273 +1 @@
-import { LlamaChatSession, } from "node-llama-cpp";
-import { SimpleChatModel } from "./base.js";
-import { createLlamaModel, createLlamaContext, } from "../util/llama_cpp.js";
-import { ChatGenerationChunk, AIMessageChunk, ChatMessage, } from "../schema/index.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.
- * @example
- * ```typescript
- * // Initialize the ChatLlamaCpp model with the path to the model binary file.
- * const model = new ChatLlamaCpp({
- *   modelPath: "/Replace/with/path/to/your/model/gguf-llama2-q4_0.bin",
- *   temperature: 0.5,
- * });
- *
- * // Call the model with a message and await the response.
- * const response = await model.call([
- *   new HumanMessage({ content: "My name is John." }),
- * ]);
- *
- * // Log the response to the console.
- * console.log({ response });
- *
- * ```
- */
-export class ChatLlamaCpp extends SimpleChatModel {
-    static lc_name() {
-        return "ChatLlamaCpp";
-    }
-    constructor(inputs) {
-        super(inputs);
-        Object.defineProperty(this, "maxTokens", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "temperature", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "topK", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "topP", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "trimWhitespaceSuffix", {
-            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.maxTokens = inputs?.maxTokens;
-        this.temperature = inputs?.temperature;
-        this.topK = inputs?.topK;
-        this.topP = inputs?.topP;
-        this.trimWhitespaceSuffix = inputs?.trimWhitespaceSuffix;
-        this._model = createLlamaModel(inputs);
-        this._context = createLlamaContext(this._model, inputs);
-        this._session = null;
-    }
-    _llmType() {
-        return "llama2_cpp";
-    }
-    /** @ignore */
-    _combineLLMOutput() {
-        return {};
-    }
-    invocationParams() {
-        return {
-            maxTokens: this.maxTokens,
-            temperature: this.temperature,
-            topK: this.topK,
-            topP: this.topP,
-            trimWhitespaceSuffix: this.trimWhitespaceSuffix,
-        };
-    }
-    /** @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 (typeof messages[0].content !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // If we already have a session then we should just have a single prompt
-            prompt = messages[0].content;
-        }
-        try {
-            const promptOptions = {
-                onToken: options.onToken,
-                maxTokens: this?.maxTokens,
-                temperature: this?.temperature,
-                topK: this?.topK,
-                topP: this?.topP,
-                trimWhitespaceSuffix: this?.trimWhitespaceSuffix,
-            };
-            // @ts-expect-error - TS2531: Object is possibly 'null'.
-            const completion = await this._session.prompt(prompt, promptOptions);
-            return completion;
-        }
-        catch (e) {
-            throw new Error("Error getting prompt completion.");
-        }
-    }
-    async *_streamResponseChunks(input, _options, runManager) {
-        const promptOptions = {
-            temperature: this?.temperature,
-            topK: this?.topK,
-            topP: this?.topP,
-        };
-        const prompt = this._buildPrompt(input);
-        const stream = await this.caller.call(async () => this._context.evaluate(this._context.encode(prompt), promptOptions));
-        for await (const chunk of stream) {
-            yield new ChatGenerationChunk({
-                text: this._context.decode([chunk]),
-                message: new AIMessageChunk({
-                    content: this._context.decode([chunk]),
-                }),
-                generationInfo: {},
-            });
-            await runManager?.handleLLMNewToken(this._context.decode([chunk]) ?? "");
-        }
-    }
-    // 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");
-            const systemMessageContent = sysMessages[sysMessages.length - 1].content;
-            if (typeof systemMessageContent !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // Only use the last provided system message
-            sysMessage = systemMessageContent;
-            // 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") {
-                const finalMessageContent = noSystemMessages[noSystemMessages.length - 1].content;
-                if (typeof finalMessageContent !== "string") {
-                    throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-                }
-                prompt = finalMessageContent;
-                interactions = this._convertMessagesToInteractions(noSystemMessages.slice(0, noSystemMessages.length - 1));
-            }
-            else {
-                interactions = this._convertMessagesToInteractions(noSystemMessages);
-            }
-        }
-        else {
-            if (typeof noSystemMessages[0].content !== "string") {
-                throw new Error("ChatLlamaCpp does not support non-string message content in sessions.");
-            }
-            // 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) {
-                const prompt = messages[i].content;
-                const response = messages[i + 1].content;
-                if (typeof prompt !== "string" || typeof response !== "string") {
-                    throw new Error("ChatLlamaCpp does not support non-string message content.");
-                }
-                result.push({
-                    prompt,
-                    response,
-                });
-            }
-        }
-        return result;
-    }
-    _buildPrompt(input) {
-        const prompt = input
-            .map((message) => {
-            let messageText;
-            if (message._getType() === "human") {
-                messageText = `[INST] ${message.content} [/INST]`;
-            }
-            else if (message._getType() === "ai") {
-                messageText = message.content;
-            }
-            else if (message._getType() === "system") {
-                messageText = `<<SYS>> ${message.content} <</SYS>>`;
-            }
-            else if (ChatMessage.isInstance(message)) {
-                messageText = `\n\n${message.role[0].toUpperCase()}${message.role.slice(1)}: ${message.content}`;
-            }
-            else {
-                console.warn(`Unsupported message type passed to llama_cpp: "${message._getType()}"`);
-                messageText = "";
-            }
-            return messageText;
-        })
-            .join("\n");
-        return prompt;
-    }
-}
+export * from "@langchain/community/chat_models/llama_cpp";