feat(anthropic,openai): Add payload formatting utils to Anthropic and…

… OpenAI (#6876)

libs/langchain-anthropic/src/index.ts

@@ -1,2 +1,2 @@
 export * from "./chat_models.js";
-export { _convertMessagesToAnthropicPayload } from "./utils/message_inputs.js";
+export { convertPromptToAnthropic } from "./utils/prompts.js";

libs/langchain-anthropic/src/tests/chat_models.test.ts

@@ -3,7 +3,7 @@ import { AIMessage, HumanMessage, ToolMessage } from "@langchain/core/messages";
 import { z } from "zod";
 import { OutputParserException } from "@langchain/core/output_parsers";
 import { ChatAnthropic } from "../chat_models.js";
-import { _convertMessagesToAnthropicPayload } from "../index.js";
+import { _convertMessagesToAnthropicPayload } from "../utils/message_inputs.js";
 
 test("withStructuredOutput with output validation", async () => {
   const model = new ChatAnthropic({

libs/langchain-anthropic/src/tests/prompts.int.test.ts

@@ -0,0 +1,28 @@
+import Anthropic from "@anthropic-ai/sdk";
+import { ChatPromptTemplate } from "@langchain/core/prompts";
+
+import { convertPromptToAnthropic } from "../utils/prompts.js";
+
+test("Convert hub prompt to Anthropic payload and invoke", async () => {
+  const prompt = ChatPromptTemplate.fromMessages([
+    ["system", "You are a world class comedian"],
+    ["human", "Tell me a joke about {topic}"],
+  ]);
+  const formattedPrompt = await prompt.invoke({
+    topic: "cats",
+  });
+
+  const { system, messages } = convertPromptToAnthropic(formattedPrompt);
+
+  const anthropicClient = new Anthropic();
+
+  const anthropicResponse = await anthropicClient.messages.create({
+    model: "claude-3-haiku-20240307",
+    system,
+    messages,
+    max_tokens: 1024,
+    stream: false,
+  });
+
+  expect(anthropicResponse.content).toBeDefined();
+});

libs/langchain-anthropic/src/utils/prompts.ts

@@ -0,0 +1,51 @@
+import type { BasePromptValue } from "@langchain/core/prompt_values";
+import Anthropic from "@anthropic-ai/sdk";
+
+import { _convertMessagesToAnthropicPayload } from "./message_inputs.js";
+
+/**
+ * Convert a formatted LangChain prompt (e.g. pulled from the hub) into
+ * a format expected by Anthropic's JS SDK.
+ *
+ * Requires the "@langchain/anthropic" package to be installed in addition
+ * to the Anthropic SDK.
+ *
+ * @example
+ * ```ts
+ * import { convertPromptToAnthropic } from "langsmith/utils/hub/anthropic";
+ * import { pull } from "langchain/hub";
+ *
+ * import Anthropic from '@anthropic-ai/sdk';
+ *
+ * const prompt = await pull("jacob/joke-generator");
+ * const formattedPrompt = await prompt.invoke({
+ *   topic: "cats",
+ * });
+ *
+ * const { system, messages } = convertPromptToAnthropic(formattedPrompt);
+ *
+ * const anthropicClient = new Anthropic({
+ *   apiKey: 'your_api_key',
+ * });
+ *
+ * const anthropicResponse = await anthropicClient.messages.create({
+ *   model: "claude-3-5-sonnet-20240620",
+ *   max_tokens: 1024,
+ *   stream: false,
+ *   system,
+ *   messages,
+ * });
+ * ```
+ * @param formattedPrompt
+ * @returns A partial Anthropic payload.
+ */
+export function convertPromptToAnthropic(
+  formattedPrompt: BasePromptValue
+): Anthropic.Messages.MessageCreateParams {
+  const messages = formattedPrompt.toChatMessages();
+  const anthropicBody = _convertMessagesToAnthropicPayload(messages);
+  if (anthropicBody.messages === undefined) {
+    anthropicBody.messages = [];
+  }
+  return anthropicBody;
+}

libs/langchain-openai/src/index.ts

@@ -9,3 +9,4 @@ export * from "./types.js";
 export * from "./utils/openai.js";
 export * from "./utils/azure.js";
 export * from "./tools/index.js";
+export { convertPromptToOpenAI } from "./utils/prompts.js";

libs/langchain-openai/src/tests/prompts.int.test.ts

@@ -0,0 +1,25 @@
+import OpenAI from "openai";
+import { ChatPromptTemplate } from "@langchain/core/prompts";
+
+import { convertPromptToOpenAI } from "../utils/prompts.js";
+
+test("Convert hub prompt to OpenAI payload and invoke", async () => {
+  const prompt = ChatPromptTemplate.fromMessages([
+    ["system", "You are a world class comedian"],
+    ["human", "Tell me a joke about {topic}"],
+  ]);
+  const formattedPrompt = await prompt.invoke({
+    topic: "cats",
+  });
+
+  const { messages } = convertPromptToOpenAI(formattedPrompt);
+
+  const openAIClient = new OpenAI();
+
+  const openAIResponse = await openAIClient.chat.completions.create({
+    model: "gpt-4o-mini",
+    messages,
+  });
+
+  expect(openAIResponse.choices.length).toBeGreaterThan(0);
+});

libs/langchain-openai/src/utils/prompts.ts

@@ -0,0 +1,47 @@
+/* eslint-disable import/no-extraneous-dependencies */
+import type { BasePromptValue } from "@langchain/core/prompt_values";
+import type { OpenAI } from "openai";
+
+import { _convertMessagesToOpenAIParams } from "../chat_models.js";
+
+/**
+ * Convert a formatted LangChain prompt (e.g. pulled from the hub) into
+ * a format expected by OpenAI's JS SDK.
+ *
+ * Requires the "@langchain/openai" package to be installed in addition
+ * to the OpenAI SDK.
+ *
+ * @example
+ * ```ts
+ * import { convertPromptToOpenAI } from "langsmith/utils/hub/openai";
+ * import { pull } from "langchain/hub";
+ *
+ * import OpenAI from 'openai';
+ *
+ * const prompt = await pull("jacob/joke-generator");
+ * const formattedPrompt = await prompt.invoke({
+ *   topic: "cats",
+ * });
+ *
+ * const { messages } = convertPromptToOpenAI(formattedPrompt);
+ *
+ * const openAIClient = new OpenAI();
+ *
+ * const openaiResponse = await openAIClient.chat.completions.create({
+ *   model: "gpt-4o",
+ *   messages,
+ * });
+ * ```
+ * @param formattedPrompt
+ * @returns A partial OpenAI payload.
+ */
+export function convertPromptToOpenAI(formattedPrompt: BasePromptValue): {
+  messages: OpenAI.Chat.ChatCompletionMessageParam[];
+} {
+  const messages = formattedPrompt.toChatMessages();
+  return {
+    messages: _convertMessagesToOpenAIParams(
+      messages
+    ) as OpenAI.Chat.ChatCompletionMessageParam[],
+  };
+}