langchain 0.0.198 → 0.0.200

package/dist/agents/index.cjs

@@ -1,9 +1,11 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.XMLAgent = exports.OpenAIAgent = exports.StructuredChatOutputParserWithRetries = exports.StructuredChatOutputParser = exports.StructuredChatAgent = exports.AgentActionOutputParser = exports.ZeroShotAgentOutputParser = exports.ZeroShotAgent = exports.initializeAgentExecutorWithOptions = exports.initializeAgentExecutor = exports.AgentExecutor = exports.ChatConversationalAgentOutputParserWithRetries = exports.ChatConversationalAgentOutputParser = exports.ChatConversationalAgent = exports.ChatAgentOutputParser = exports.ChatAgent = exports.Toolkit = exports.createVectorStoreRouterAgent = exports.createVectorStoreAgent = exports.createOpenApiAgent = exports.createJsonAgent = exports.ZapierToolKit = exports.VectorStoreToolkit = exports.VectorStoreRouterToolkit = exports.RequestsToolkit = exports.OpenApiToolkit = exports.JsonToolkit = exports.LLMSingleActionAgent = exports.BaseSingleActionAgent = exports.Agent = void 0;
+exports.XMLAgent = exports.OpenAIAgent = exports.StructuredChatOutputParserWithRetries = exports.StructuredChatOutputParser = exports.StructuredChatAgent = exports.AgentActionOutputParser = exports.ZeroShotAgentOutputParser = exports.ZeroShotAgent = exports.initializeAgentExecutorWithOptions = exports.initializeAgentExecutor = exports.AgentExecutor = exports.ChatConversationalAgentOutputParserWithRetries = exports.ChatConversationalAgentOutputParser = exports.ChatConversationalAgent = exports.ChatAgentOutputParser = exports.ChatAgent = exports.Toolkit = exports.createVectorStoreRouterAgent = exports.createVectorStoreAgent = exports.createOpenApiAgent = exports.createJsonAgent = exports.ZapierToolKit = exports.VectorStoreToolkit = exports.VectorStoreRouterToolkit = exports.RequestsToolkit = exports.OpenApiToolkit = exports.JsonToolkit = exports.LLMSingleActionAgent = exports.RunnableAgent = exports.BaseMultiActionAgent = exports.BaseSingleActionAgent = exports.Agent = void 0;
 var agent_js_1 = require("./agent.cjs");
 Object.defineProperty(exports, "Agent", { enumerable: true, get: function () { return agent_js_1.Agent; } });
 Object.defineProperty(exports, "BaseSingleActionAgent", { enumerable: true, get: function () { return agent_js_1.BaseSingleActionAgent; } });
+Object.defineProperty(exports, "BaseMultiActionAgent", { enumerable: true, get: function () { return agent_js_1.BaseMultiActionAgent; } });
+Object.defineProperty(exports, "RunnableAgent", { enumerable: true, get: function () { return agent_js_1.RunnableAgent; } });
 Object.defineProperty(exports, "LLMSingleActionAgent", { enumerable: true, get: function () { return agent_js_1.LLMSingleActionAgent; } });
 var index_js_1 = require("./toolkits/index.cjs");
 Object.defineProperty(exports, "JsonToolkit", { enumerable: true, get: function () { return index_js_1.JsonToolkit; } });

package/dist/agents/index.d.ts

@@ -1,4 +1,4 @@
-export { Agent, type AgentArgs, BaseSingleActionAgent, LLMSingleActionAgent, type LLMSingleActionAgentInput, type OutputParserArgs, } from "./agent.js";
+export { Agent, type AgentArgs, BaseSingleActionAgent, BaseMultiActionAgent, RunnableAgent, LLMSingleActionAgent, type LLMSingleActionAgentInput, type OutputParserArgs, } from "./agent.js";
 export { JsonToolkit, OpenApiToolkit, RequestsToolkit, type VectorStoreInfo, VectorStoreRouterToolkit, VectorStoreToolkit, ZapierToolKit, createJsonAgent, createOpenApiAgent, createVectorStoreAgent, createVectorStoreRouterAgent, } from "./toolkits/index.js";
 export { Toolkit } from "./toolkits/base.js";
 export { ChatAgent, type ChatAgentInput, type ChatCreatePromptArgs, } from "./chat/index.js";
@@ -6,7 +6,7 @@ export { ChatAgentOutputParser } from "./chat/outputParser.js";
 export { ChatConversationalAgent, type ChatConversationalAgentInput, type ChatConversationalCreatePromptArgs, } from "./chat_convo/index.js";
 export { ChatConversationalAgentOutputParser, type ChatConversationalAgentOutputParserArgs, ChatConversationalAgentOutputParserWithRetries, type ChatConversationalAgentOutputParserFormatInstructionsOptions, } from "./chat_convo/outputParser.js";
 export { AgentExecutor, type AgentExecutorInput } from "./executor.js";
-export { initializeAgentExecutor, initializeAgentExecutorWithOptions, type InitializeAgentExecutorOptions, } from "./initialize.js";
+export { initializeAgentExecutor, initializeAgentExecutorWithOptions, type InitializeAgentExecutorOptions, type InitializeAgentExecutorOptionsStructured, } from "./initialize.js";
 export { ZeroShotAgent, type ZeroShotAgentInput, type ZeroShotCreatePromptArgs, } from "./mrkl/index.js";
 export { ZeroShotAgentOutputParser } from "./mrkl/outputParser.js";
 export { AgentActionOutputParser, type AgentInput, type SerializedAgent, type SerializedAgentT, type SerializedZeroShotAgent, type StoppingMethod, } from "./types.js";

package/dist/agents/index.js

@@ -1,4 +1,4 @@
-export { Agent, BaseSingleActionAgent, LLMSingleActionAgent, } from "./agent.js";
+export { Agent, BaseSingleActionAgent, BaseMultiActionAgent, RunnableAgent, LLMSingleActionAgent, } from "./agent.js";
 export { JsonToolkit, OpenApiToolkit, RequestsToolkit, VectorStoreRouterToolkit, VectorStoreToolkit, ZapierToolKit, createJsonAgent, createOpenApiAgent, createVectorStoreAgent, createVectorStoreRouterAgent, } from "./toolkits/index.js";
 export { Toolkit } from "./toolkits/base.js";
 export { ChatAgent, } from "./chat/index.js";

package/dist/agents/toolkits/conversational_retrieval/tool.cjs

@@ -7,7 +7,7 @@ const document_js_1 = require("../../../util/document.cjs");
 function createRetrieverTool(retriever, input) {
     const func = async ({ input }, runManager) => {
         const docs = await retriever.getRelevantDocuments(input, runManager?.getChild("retriever"));
-        return (0, document_js_1.formatDocumentsAsString)(docs, "\n");
+        return (0, document_js_1.formatDocumentsAsString)(docs);
     };
     const schema = zod_1.z.object({
         input: zod_1.z

package/dist/agents/toolkits/conversational_retrieval/tool.js

@@ -4,7 +4,7 @@ import { formatDocumentsAsString } from "../../../util/document.js";
 export function createRetrieverTool(retriever, input) {
     const func = async ({ input }, runManager) => {
         const docs = await retriever.getRelevantDocuments(input, runManager?.getChild("retriever"));
-        return formatDocumentsAsString(docs, "\n");
+        return formatDocumentsAsString(docs);
     };
     const schema = z.object({
         input: z

package/dist/chains/conversational_retrieval_chain.cjs

@@ -94,12 +94,20 @@ class ConversationalRetrievalQAChain extends base_js_1.BaseChain {
             writable: true,
             value: false
         });
+        Object.defineProperty(this, "returnGeneratedQuestion", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: false
+        });
         this.retriever = fields.retriever;
         this.combineDocumentsChain = fields.combineDocumentsChain;
         this.questionGeneratorChain = fields.questionGeneratorChain;
         this.inputKey = fields.inputKey ?? this.inputKey;
         this.returnSourceDocuments =
             fields.returnSourceDocuments ?? this.returnSourceDocuments;
+        this.returnGeneratedQuestion =
+            fields.returnGeneratedQuestion ?? this.returnGeneratedQuestion;
     }
     /**
      * Static method to convert the chat history input into a formatted
@@ -172,13 +180,19 @@ class ConversationalRetrievalQAChain extends base_js_1.BaseChain {
             input_documents: docs,
             chat_history: chatHistory,
         };
-        const result = await this.combineDocumentsChain.call(inputs, runManager?.getChild("combine_documents"));
+        let result = await this.combineDocumentsChain.call(inputs, runManager?.getChild("combine_documents"));
         if (this.returnSourceDocuments) {
-            return {
+            result = {
                 ...result,
                 sourceDocuments: docs,
             };
         }
+        if (this.returnGeneratedQuestion) {
+            result = {
+                ...result,
+                generatedQuestion: newQuestion,
+            };
+        }
         return result;
     }
     _chainType() {

package/dist/chains/conversational_retrieval_chain.d.ts

@@ -16,6 +16,7 @@ export interface ConversationalRetrievalQAChainInput extends ChainInputs {
     combineDocumentsChain: BaseChain;
     questionGeneratorChain: LLMChain;
     returnSourceDocuments?: boolean;
+    returnGeneratedQuestion?: boolean;
     inputKey?: string;
 }
 /**
@@ -62,6 +63,7 @@ export declare class ConversationalRetrievalQAChain extends BaseChain implements
     combineDocumentsChain: BaseChain;
     questionGeneratorChain: LLMChain;
     returnSourceDocuments: boolean;
+    returnGeneratedQuestion: boolean;
     constructor(fields: ConversationalRetrievalQAChainInput);
     /**
      * Static method to convert the chat history input into a formatted

package/dist/chains/conversational_retrieval_chain.js

@@ -91,12 +91,20 @@ export class ConversationalRetrievalQAChain extends BaseChain {
             writable: true,
             value: false
         });
+        Object.defineProperty(this, "returnGeneratedQuestion", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: false
+        });
         this.retriever = fields.retriever;
         this.combineDocumentsChain = fields.combineDocumentsChain;
         this.questionGeneratorChain = fields.questionGeneratorChain;
         this.inputKey = fields.inputKey ?? this.inputKey;
         this.returnSourceDocuments =
             fields.returnSourceDocuments ?? this.returnSourceDocuments;
+        this.returnGeneratedQuestion =
+            fields.returnGeneratedQuestion ?? this.returnGeneratedQuestion;
     }
     /**
      * Static method to convert the chat history input into a formatted
@@ -169,13 +177,19 @@ export class ConversationalRetrievalQAChain extends BaseChain {
             input_documents: docs,
             chat_history: chatHistory,
         };
-        const result = await this.combineDocumentsChain.call(inputs, runManager?.getChild("combine_documents"));
+        let result = await this.combineDocumentsChain.call(inputs, runManager?.getChild("combine_documents"));
         if (this.returnSourceDocuments) {
-            return {
+            result = {
                 ...result,
                 sourceDocuments: docs,
             };
         }
+        if (this.returnGeneratedQuestion) {
+            result = {
+                ...result,
+                generatedQuestion: newQuestion,
+            };
+        }
         return result;
     }
     _chainType() {

package/dist/chat_models/fake.cjs

@@ -1,117 +1,5 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.FakeListChatModel = void 0;
-const base_js_1 = require("./base.cjs");
-const index_js_1 = require("../schema/index.cjs");
-/**
- * A fake Chat Model that returns a predefined list of responses. It can be used
- * for testing purposes.
- * @example
- * ```typescript
- * const chat = new FakeListChatModel({
- *   responses: ["I'll callback later.", "You 'console' them!"]
- * });
- *
- * const firstMessage = new HumanMessage("You want to hear a JavaScript joke?");
- * const secondMessage = new HumanMessage("How do you cheer up a JavaScript developer?");
- *
- * // Call the chat model with a message and log the response
- * const firstResponse = await chat.call([firstMessage]);
- * console.log({ firstResponse });
- *
- * const secondResponse = await chat.call([secondMessage]);
- * console.log({ secondResponse });
- * ```
- */
-class FakeListChatModel extends base_js_1.BaseChatModel {
-    static lc_name() {
-        return "FakeListChatModel";
-    }
-    constructor({ responses, sleep }) {
-        super({});
-        Object.defineProperty(this, "responses", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "i", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: 0
-        });
-        Object.defineProperty(this, "sleep", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        this.responses = responses;
-        this.sleep = sleep;
-    }
-    _combineLLMOutput() {
-        return [];
-    }
-    _llmType() {
-        return "fake-list";
-    }
-    async _generate(_messages, options) {
-        await this._sleepIfRequested();
-        if (options?.stop?.length) {
-            return {
-                generations: [this._formatGeneration(options.stop[0])],
-            };
-        }
-        else {
-            const response = this._currentResponse();
-            this._incrementResponse();
-            return {
-                generations: [this._formatGeneration(response)],
-                llmOutput: {},
-            };
-        }
-    }
-    _formatGeneration(text) {
-        return {
-            message: new index_js_1.AIMessage(text),
-            text,
-        };
-    }
-    async *_streamResponseChunks(_messages, _options, _runManager) {
-        const response = this._currentResponse();
-        this._incrementResponse();
-        for await (const text of response) {
-            await this._sleepIfRequested();
-            yield this._createResponseChunk(text);
-        }
-    }
-    async _sleepIfRequested() {
-        if (this.sleep !== undefined) {
-            await this._sleep();
-        }
-    }
-    async _sleep() {
-        return new Promise((resolve) => {
-            setTimeout(() => resolve(), this.sleep);
-        });
-    }
-    _createResponseChunk(text) {
-        return new index_js_1.ChatGenerationChunk({
-            message: new index_js_1.AIMessageChunk({ content: text }),
-            text,
-        });
-    }
-    _currentResponse() {
-        return this.responses[this.i];
-    }
-    _incrementResponse() {
-        if (this.i < this.responses.length - 1) {
-            this.i += 1;
-        }
-        else {
-            this.i = 0;
-        }
-    }
-}
-exports.FakeListChatModel = FakeListChatModel;
+var testing_1 = require("@langchain/core/utils/testing");
+Object.defineProperty(exports, "FakeListChatModel", { enumerable: true, get: function () { return testing_1.FakeListChatModel; } });

package/dist/chat_models/fake.d.ts

@@ -1,52 +1 @@
-import { BaseChatModel, BaseChatModelParams } from "./base.js";
-import { AIMessage, BaseMessage, ChatGenerationChunk, ChatResult } from "../schema/index.js";
-import { CallbackManagerForLLMRun } from "../callbacks/manager.js";
-/**
- * Interface for the input parameters specific to the Fake List Chat model.
- */
-export interface FakeChatInput extends BaseChatModelParams {
-    /** Responses to return */
-    responses: string[];
-    /** Time to sleep in milliseconds between responses */
-    sleep?: number;
-}
-/**
- * A fake Chat Model that returns a predefined list of responses. It can be used
- * for testing purposes.
- * @example
- * ```typescript
- * const chat = new FakeListChatModel({
- *   responses: ["I'll callback later.", "You 'console' them!"]
- * });
- *
- * const firstMessage = new HumanMessage("You want to hear a JavaScript joke?");
- * const secondMessage = new HumanMessage("How do you cheer up a JavaScript developer?");
- *
- * // Call the chat model with a message and log the response
- * const firstResponse = await chat.call([firstMessage]);
- * console.log({ firstResponse });
- *
- * const secondResponse = await chat.call([secondMessage]);
- * console.log({ secondResponse });
- * ```
- */
-export declare class FakeListChatModel extends BaseChatModel {
-    static lc_name(): string;
-    responses: string[];
-    i: number;
-    sleep?: number;
-    constructor({ responses, sleep }: FakeChatInput);
-    _combineLLMOutput(): never[];
-    _llmType(): string;
-    _generate(_messages: BaseMessage[], options?: this["ParsedCallOptions"]): Promise<ChatResult>;
-    _formatGeneration(text: string): {
-        message: AIMessage;
-        text: string;
-    };
-    _streamResponseChunks(_messages: BaseMessage[], _options: this["ParsedCallOptions"], _runManager?: CallbackManagerForLLMRun): AsyncGenerator<ChatGenerationChunk>;
-    _sleepIfRequested(): Promise<void>;
-    _sleep(): Promise<void>;
-    _createResponseChunk(text: string): ChatGenerationChunk;
-    _currentResponse(): string;
-    _incrementResponse(): void;
-}
+export { type FakeChatInput, FakeListChatModel, } from "@langchain/core/utils/testing";

package/dist/chat_models/fake.js

@@ -1,113 +1 @@
-import { BaseChatModel } from "./base.js";
-import { AIMessage, AIMessageChunk, ChatGenerationChunk, } from "../schema/index.js";
-/**
- * A fake Chat Model that returns a predefined list of responses. It can be used
- * for testing purposes.
- * @example
- * ```typescript
- * const chat = new FakeListChatModel({
- *   responses: ["I'll callback later.", "You 'console' them!"]
- * });
- *
- * const firstMessage = new HumanMessage("You want to hear a JavaScript joke?");
- * const secondMessage = new HumanMessage("How do you cheer up a JavaScript developer?");
- *
- * // Call the chat model with a message and log the response
- * const firstResponse = await chat.call([firstMessage]);
- * console.log({ firstResponse });
- *
- * const secondResponse = await chat.call([secondMessage]);
- * console.log({ secondResponse });
- * ```
- */
-export class FakeListChatModel extends BaseChatModel {
-    static lc_name() {
-        return "FakeListChatModel";
-    }
-    constructor({ responses, sleep }) {
-        super({});
-        Object.defineProperty(this, "responses", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "i", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: 0
-        });
-        Object.defineProperty(this, "sleep", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        this.responses = responses;
-        this.sleep = sleep;
-    }
-    _combineLLMOutput() {
-        return [];
-    }
-    _llmType() {
-        return "fake-list";
-    }
-    async _generate(_messages, options) {
-        await this._sleepIfRequested();
-        if (options?.stop?.length) {
-            return {
-                generations: [this._formatGeneration(options.stop[0])],
-            };
-        }
-        else {
-            const response = this._currentResponse();
-            this._incrementResponse();
-            return {
-                generations: [this._formatGeneration(response)],
-                llmOutput: {},
-            };
-        }
-    }
-    _formatGeneration(text) {
-        return {
-            message: new AIMessage(text),
-            text,
-        };
-    }
-    async *_streamResponseChunks(_messages, _options, _runManager) {
-        const response = this._currentResponse();
-        this._incrementResponse();
-        for await (const text of response) {
-            await this._sleepIfRequested();
-            yield this._createResponseChunk(text);
-        }
-    }
-    async _sleepIfRequested() {
-        if (this.sleep !== undefined) {
-            await this._sleep();
-        }
-    }
-    async _sleep() {
-        return new Promise((resolve) => {
-            setTimeout(() => resolve(), this.sleep);
-        });
-    }
-    _createResponseChunk(text) {
-        return new ChatGenerationChunk({
-            message: new AIMessageChunk({ content: text }),
-            text,
-        });
-    }
-    _currentResponse() {
-        return this.responses[this.i];
-    }
-    _incrementResponse() {
-        if (this.i < this.responses.length - 1) {
-            this.i += 1;
-        }
-        else {
-            this.i = 0;
-        }
-    }
-}
+export { FakeListChatModel, } from "@langchain/core/utils/testing";

package/dist/chat_models/llama_cpp.cjs

@@ -108,7 +108,7 @@ class ChatLlamaCpp extends base_js_1.SimpleChatModel {
         };
     }
     /** @ignore */
-    async _call(messages, _options) {
+    async _call(messages, options) {
         let prompt = "";
         if (messages.length > 1) {
             // We need to build a new _session
@@ -126,6 +126,7 @@ class ChatLlamaCpp extends base_js_1.SimpleChatModel {
         }
         try {
             const promptOptions = {
+                onToken: options.onToken,
                 maxTokens: this?.maxTokens,
                 temperature: this?.temperature,
                 topK: this?.topK,
@@ -141,26 +142,22 @@ class ChatLlamaCpp extends base_js_1.SimpleChatModel {
         }
     }
     async *_streamResponseChunks(input, _options, runManager) {
-        if (input.length !== 1) {
-            throw new Error("Only one human message should be provided.");
-        }
-        else {
-            const promptOptions = {
-                temperature: this?.temperature,
-                topK: this?.topK,
-                topP: this?.topP,
-            };
-            const stream = await this.caller.call(async () => this._context.evaluate(this._context.encode(`${input[0].content}`), 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]) ?? "");
-            }
+        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
@@ -251,5 +248,30 @@ class ChatLlamaCpp extends base_js_1.SimpleChatModel {
         }
         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;

package/dist/chat_models/llama_cpp.d.ts

@@ -63,8 +63,9 @@ export declare class ChatLlamaCpp extends SimpleChatModel<LlamaCppCallOptions> {
         trimWhitespaceSuffix: boolean | undefined;
     };
     /** @ignore */
-    _call(messages: BaseMessage[], _options: this["ParsedCallOptions"]): Promise<string>;
+    _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;
 }

package/dist/chat_models/llama_cpp.js

@@ -1,7 +1,7 @@
 import { LlamaChatSession, } from "node-llama-cpp";
 import { SimpleChatModel } from "./base.js";
 import { createLlamaModel, createLlamaContext, } from "../util/llama_cpp.js";
-import { ChatGenerationChunk, AIMessageChunk, } from "../schema/index.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
@@ -105,7 +105,7 @@ export class ChatLlamaCpp extends SimpleChatModel {
         };
     }
     /** @ignore */
-    async _call(messages, _options) {
+    async _call(messages, options) {
         let prompt = "";
         if (messages.length > 1) {
             // We need to build a new _session
@@ -123,6 +123,7 @@ export class ChatLlamaCpp extends SimpleChatModel {
         }
         try {
             const promptOptions = {
+                onToken: options.onToken,
                 maxTokens: this?.maxTokens,
                 temperature: this?.temperature,
                 topK: this?.topK,
@@ -138,26 +139,22 @@ export class ChatLlamaCpp extends SimpleChatModel {
         }
     }
     async *_streamResponseChunks(input, _options, runManager) {
-        if (input.length !== 1) {
-            throw new Error("Only one human message should be provided.");
-        }
-        else {
-            const promptOptions = {
-                temperature: this?.temperature,
-                topK: this?.topK,
-                topP: this?.topP,
-            };
-            const stream = await this.caller.call(async () => this._context.evaluate(this._context.encode(`${input[0].content}`), 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]) ?? "");
-            }
+        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
@@ -248,4 +245,29 @@ export class ChatLlamaCpp extends SimpleChatModel {
         }
         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;
+    }
 }

package/dist/chat_models/minimax.d.ts

@@ -1,4 +1,4 @@
-import type { OpenAIClient } from "@langchain/openai";
+import type { OpenAI as OpenAIClient } from "openai";
 import { BaseChatModel, BaseChatModelParams } from "./base.js";
 import { BaseMessage, ChatResult } from "../schema/index.js";
 import { CallbackManagerForLLMRun } from "../callbacks/manager.js";