anthropic_functions.ts

import { XMLParser } from "fast-xml-parser";
import type { StructuredToolInterface } from "@langchain/core/tools";

import {
  AIMessage,
  BaseMessage,
  SystemMessage,
} from "@langchain/core/messages";
import { ChatGenerationChunk, ChatResult } from "@langchain/core/outputs";
import {
  BaseChatModel,
  BaseChatModelParams,
} from "@langchain/core/language_models/chat_models";
import { BaseFunctionCallOptions } from "@langchain/core/language_models/base";
import { CallbackManagerForLLMRun } from "@langchain/core/callbacks/manager";
import { BasePromptTemplate, PromptTemplate } from "@langchain/core/prompts";
import { convertToOpenAIFunction } from "@langchain/core/utils/function_calling";
import {
  ChatAnthropic,
  DEFAULT_STOP_SEQUENCES,
  type AnthropicInput,
} from "../../chat_models/anthropic.js";

const TOOL_SYSTEM_PROMPT =
  /* #__PURE__ */
  PromptTemplate.fromTemplate(`In addition to responding, you can use tools.
You should use tools as often as you can, as they return the most accurate information possible.
You have access to the following tools:

{tools}

In order to use a tool, you can use <tool></tool> to specify the name,
and the <tool_input></tool_input> tags to specify the parameters.
Each parameter should be passed in as <$param_name>$value</$param_name>,
Where $param_name is the name of the specific parameter, and $value
is the value for that parameter.

You will then get back a response in the form <observation></observation>
For example, if you have a tool called 'search' that accepts a single
parameter 'query' that could run a google search, in order to search
for the weather in SF you would respond:

<tool>search</tool><tool_input><query>weather in SF</query></tool_input>
<observation>64 degrees</observation>`);

/** @deprecated Install and use in "@langchain/anthropic/experimental" instead */
export interface ChatAnthropicFunctionsCallOptions
  extends BaseFunctionCallOptions {
  tools?: StructuredToolInterface[];
}

/** @deprecated Install and use in "@langchain/anthropic/experimental" instead */
export type AnthropicFunctionsInput = Partial<AnthropicInput> &
  BaseChatModelParams & {
    llm?: BaseChatModel;
    systemPromptTemplate?: BasePromptTemplate;
  };

/** @deprecated Install and use in "@langchain/anthropic/experimental" instead */
export class AnthropicFunctions extends BaseChatModel<ChatAnthropicFunctionsCallOptions> {
  llm: BaseChatModel;

  stopSequences?: string[];

  systemPromptTemplate: BasePromptTemplate;

  lc_namespace = ["langchain", "experimental", "chat_models"];

  static lc_name(): string {
    return "AnthropicFunctions";
  }

  constructor(fields?: AnthropicFunctionsInput) {
    super(fields ?? {});
    this.llm = fields?.llm ?? new ChatAnthropic(fields);
    this.systemPromptTemplate =
      fields?.systemPromptTemplate ?? TOOL_SYSTEM_PROMPT;
    this.stopSequences =
      fields?.stopSequences ?? (this.llm as ChatAnthropic).stopSequences;
  }

  invocationParams() {
    return this.llm.invocationParams();
  }

  /** @ignore */
  _identifyingParams() {
    return this.llm._identifyingParams();
  }

  async *_streamResponseChunks(
    messages: BaseMessage[],
    options: this["ParsedCallOptions"],
    runManager?: CallbackManagerForLLMRun
  ): AsyncGenerator<ChatGenerationChunk> {
    yield* this.llm._streamResponseChunks(messages, options, runManager);
  }

  async _generate(
    messages: BaseMessage[],
    options: this["ParsedCallOptions"],
    runManager?: CallbackManagerForLLMRun | undefined
  ): Promise<ChatResult> {
    let promptMessages = messages;
    let forced = false;
    let functionCall: string | undefined;
    if (options.tools) {
      // eslint-disable-next-line no-param-reassign
      options.functions = (options.functions ?? []).concat(
        options.tools.map(convertToOpenAIFunction)
      );
    }
    if (options.functions !== undefined && options.functions.length > 0) {
      const content = await this.systemPromptTemplate.format({
        tools: JSON.stringify(options.functions, null, 2),
      });
      const systemMessage = new SystemMessage({ content });
      promptMessages = [systemMessage].concat(promptMessages);
      const stopSequences =
        options?.stop?.concat(DEFAULT_STOP_SEQUENCES) ??
        this.stopSequences ??
        DEFAULT_STOP_SEQUENCES;
      // eslint-disable-next-line no-param-reassign
      options.stop = stopSequences.concat(["</tool_input>"]);
      if (options.function_call) {
        if (typeof options.function_call === "string") {
          functionCall = JSON.parse(options.function_call).name;
        } else {
          functionCall = options.function_call.name;
        }
        forced = true;
        const matchingFunction = options.functions.find(
          (tool) => tool.name === functionCall
        );
        if (!matchingFunction) {
          throw new Error(
            `No matching function found for passed "function_call"`
          );
        }
        promptMessages = promptMessages.concat([
          new AIMessage({
            content: `<tool>${functionCall}</tool>`,
          }),
        ]);
        // eslint-disable-next-line no-param-reassign
        delete options.function_call;
      }
      // eslint-disable-next-line no-param-reassign
      delete options.functions;
    } else if (options.function_call !== undefined) {
      throw new Error(
        `If "function_call" is provided, "functions" must also be.`
      );
    }
    const chatResult = await this.llm._generate(
      promptMessages,
      options,
      runManager
    );
    const chatGenerationContent = chatResult.generations[0].message.content;
    if (typeof chatGenerationContent !== "string") {
      throw new Error("AnthropicFunctions does not support non-string output.");
    }

    if (forced) {
      const parser = new XMLParser();
      const result = parser.parse(`${chatGenerationContent}</tool_input>`);
      if (functionCall === undefined) {
        throw new Error(`Could not parse called function from model output.`);
      }
      const responseMessageWithFunctions = new AIMessage({
        content: "",
        additional_kwargs: {
          function_call: {
            name: functionCall,
            arguments: result.tool_input
              ? JSON.stringify(result.tool_input)
              : "",
          },
        },
      });
      return {
        generations: [{ message: responseMessageWithFunctions, text: "" }],
      };
    } else if (chatGenerationContent.includes("<tool>")) {
      const parser = new XMLParser();
      const result = parser.parse(`${chatGenerationContent}</tool_input>`);
      const responseMessageWithFunctions = new AIMessage({
        content: chatGenerationContent.split("<tool>")[0],
        additional_kwargs: {
          function_call: {
            name: result.tool,
            arguments: result.tool_input
              ? JSON.stringify(result.tool_input)
              : "",
          },
        },
      });
      return {
        generations: [{ message: responseMessageWithFunctions, text: "" }],
      };
    }
    return chatResult;
  }

  _llmType(): string {
    return "anthropic_functions";
  }

  /** @ignore */
  _combineLLMOutput() {
    return [];
  }
}