@langchain/openai 0.2.6 → 0.2.8

package/dist/chat_models.cjs

@@ -7,14 +7,15 @@ const outputs_1 = require("@langchain/core/outputs");
 const env_1 = require("@langchain/core/utils/env");
 const chat_models_1 = require("@langchain/core/language_models/chat_models");
 const base_1 = require("@langchain/core/language_models/base");
-const function_calling_1 = require("@langchain/core/utils/function_calling");
 const runnables_1 = require("@langchain/core/runnables");
 const output_parsers_1 = require("@langchain/core/output_parsers");
 const openai_tools_1 = require("@langchain/core/output_parsers/openai_tools");
 const zod_to_json_schema_1 = require("zod-to-json-schema");
+const zod_1 = require("openai/helpers/zod");
 const azure_js_1 = require("./utils/azure.cjs");
 const openai_js_1 = require("./utils/openai.cjs");
 const openai_format_fndef_js_1 = require("./utils/openai-format-fndef.cjs");
+const tools_js_1 = require("./utils/tools.cjs");
 function extractGenericMessageCustomRole(message) {
     if (message.role !== "system" &&
         message.role !== "assistant" &&
@@ -198,40 +199,466 @@ function _convertChatOpenAIToolTypeToOpenAITool(tool, fields) {
         }
         return tool;
     }
-    return (0, function_calling_1.convertToOpenAITool)(tool, fields);
+    return (0, tools_js_1._convertToOpenAITool)(tool, fields);
 }
 /**
- * Wrapper around OpenAI large language models that use the Chat endpoint.
- *
- * To use you should have the `OPENAI_API_KEY` environment variable set.
- *
- * To use with Azure you should have the:
- * `AZURE_OPENAI_API_KEY`,
- * `AZURE_OPENAI_API_INSTANCE_NAME`,
- * `AZURE_OPENAI_API_DEPLOYMENT_NAME`
- * and `AZURE_OPENAI_API_VERSION` environment variables set.
- * `AZURE_OPENAI_BASE_PATH` is optional and will override `AZURE_OPENAI_API_INSTANCE_NAME` if you need to use a custom endpoint.
- *
- * @remarks
- * Any parameters that are valid to be passed to {@link
- * https://platform.openai.com/docs/api-reference/chat/create |
- * `openai.createChatCompletion`} can be passed through {@link modelKwargs}, even
- * if not explicitly available on this class.
- * @example
+ * OpenAI chat model integration.
+ *
+ * Setup:
+ * Install `@langchain/openai` and set an environment variable named `OPENAI_API_KEY`.
+ *
+ * ```bash
+ * npm install @langchain/openai
+ * export OPENAI_API_KEY="your-api-key"
+ * ```
+ *
+ * ## [Constructor args](https://api.js.langchain.com/classes/langchain_openai.ChatOpenAI.html#constructor)
+ *
+ * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_openai.ChatOpenAICallOptions.html)
+ *
+ * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.
+ * They can also be passed via `.bind`, or the second arg in `.bindTools`, like shown in the examples below:
+ *
+ * ```typescript
+ * // When calling `.bind`, call options should be passed via the first argument
+ * const llmWithArgsBound = llm.bind({
+ *   stop: ["\n"],
+ *   tools: [...],
+ * });
+ *
+ * // When calling `.bindTools`, call options should be passed via the second argument
+ * const llmWithTools = llm.bindTools(
+ *   [...],
+ *   {
+ *     tool_choice: "auto",
+ *   }
+ * );
+ * ```
+ *
+ * ## Examples
+ *
+ * <details open>
+ * <summary><strong>Instantiate</strong></summary>
+ *
  * ```typescript
- * // Create a new instance of ChatOpenAI with specific temperature and model name settings
- * const model = new ChatOpenAI({
- *   temperature: 0.9,
- *   model: "ft:gpt-3.5-turbo-0613:{ORG_NAME}::{MODEL_ID}",
+ * import { ChatOpenAI } from '@langchain/openai';
+ *
+ * const llm = new ChatOpenAI({
+ *   model: "gpt-4o",
+ *   temperature: 0,
+ *   maxTokens: undefined,
+ *   timeout: undefined,
+ *   maxRetries: 2,
+ *   // apiKey: "...",
+ *   // baseUrl: "...",
+ *   // organization: "...",
+ *   // other params...
  * });
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Invoking</strong></summary>
+ *
+ * ```typescript
+ * const input = `Translate "I love programming" into French.`;
+ *
+ * // Models also accept a list of chat messages or a formatted prompt
+ * const result = await llm.invoke(input);
+ * console.log(result);
+ * ```
+ *
+ * ```txt
+ * AIMessage {
+ *   "id": "chatcmpl-9u4Mpu44CbPjwYFkTbeoZgvzB00Tz",
+ *   "content": "J'adore la programmation.",
+ *   "response_metadata": {
+ *     "tokenUsage": {
+ *       "completionTokens": 5,
+ *       "promptTokens": 28,
+ *       "totalTokens": 33
+ *     },
+ *     "finish_reason": "stop",
+ *     "system_fingerprint": "fp_3aa7262c27"
+ *   },
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Streaming Chunks</strong></summary>
+ *
+ * ```typescript
+ * for await (const chunk of await llm.stream(input)) {
+ *   console.log(chunk);
+ * }
+ * ```
+ *
+ * ```txt
+ * AIMessageChunk {
+ *   "id": "chatcmpl-9u4NWB7yUeHCKdLr6jP3HpaOYHTqs",
+ *   "content": ""
+ * }
+ * AIMessageChunk {
+ *   "content": "J"
+ * }
+ * AIMessageChunk {
+ *   "content": "'adore"
+ * }
+ * AIMessageChunk {
+ *   "content": " la"
+ * }
+ * AIMessageChunk {
+ *   "content": " programmation",,
+ * }
+ * AIMessageChunk {
+ *   "content": ".",,
+ * }
+ * AIMessageChunk {
+ *   "content": "",
+ *   "response_metadata": {
+ *     "finish_reason": "stop",
+ *     "system_fingerprint": "fp_c9aa9c0491"
+ *   },
+ * }
+ * AIMessageChunk {
+ *   "content": "",
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Aggregate Streamed Chunks</strong></summary>
+ *
+ * ```typescript
+ * import { AIMessageChunk } from '@langchain/core/messages';
+ * import { concat } from '@langchain/core/utils/stream';
  *
- * // Invoke the model with a message and await the response
- * const message = await model.invoke("Hi there!");
+ * const stream = await llm.stream(input);
+ * let full: AIMessageChunk | undefined;
+ * for await (const chunk of stream) {
+ *   full = !full ? chunk : concat(full, chunk);
+ * }
+ * console.log(full);
+ * ```
+ *
+ * ```txt
+ * AIMessageChunk {
+ *   "id": "chatcmpl-9u4PnX6Fy7OmK46DASy0bH6cxn5Xu",
+ *   "content": "J'adore la programmation.",
+ *   "response_metadata": {
+ *     "prompt": 0,
+ *     "completion": 0,
+ *     "finish_reason": "stop",
+ *   },
+ *   "usage_metadata": {
+ *     "input_tokens": 28,
+ *     "output_tokens": 5,
+ *     "total_tokens": 33
+ *   }
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Bind tools</strong></summary>
+ *
+ * ```typescript
+ * import { z } from 'zod';
+ *
+ * const GetWeather = {
+ *   name: "GetWeather",
+ *   description: "Get the current weather in a given location",
+ *   schema: z.object({
+ *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
+ *   }),
+ * }
+ *
+ * const GetPopulation = {
+ *   name: "GetPopulation",
+ *   description: "Get the current population in a given location",
+ *   schema: z.object({
+ *     location: z.string().describe("The city and state, e.g. San Francisco, CA")
+ *   }),
+ * }
+ *
+ * const llmWithTools = llm.bindTools(
+ *   [GetWeather, GetPopulation],
+ *   {
+ *     // strict: true  // enforce tool args schema is respected
+ *   }
+ * );
+ * const aiMsg = await llmWithTools.invoke(
+ *   "Which city is hotter today and which is bigger: LA or NY?"
+ * );
+ * console.log(aiMsg.tool_calls);
+ * ```
+ *
+ * ```txt
+ * [
+ *   {
+ *     name: 'GetWeather',
+ *     args: { location: 'Los Angeles, CA' },
+ *     type: 'tool_call',
+ *     id: 'call_uPU4FiFzoKAtMxfmPnfQL6UK'
+ *   },
+ *   {
+ *     name: 'GetWeather',
+ *     args: { location: 'New York, NY' },
+ *     type: 'tool_call',
+ *     id: 'call_UNkEwuQsHrGYqgDQuH9nPAtX'
+ *   },
+ *   {
+ *     name: 'GetPopulation',
+ *     args: { location: 'Los Angeles, CA' },
+ *     type: 'tool_call',
+ *     id: 'call_kL3OXxaq9OjIKqRTpvjaCH14'
+ *   },
+ *   {
+ *     name: 'GetPopulation',
+ *     args: { location: 'New York, NY' },
+ *     type: 'tool_call',
+ *     id: 'call_s9KQB1UWj45LLGaEnjz0179q'
+ *   }
+ * ]
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Structured Output</strong></summary>
+ *
+ * ```typescript
+ * import { z } from 'zod';
+ *
+ * const Joke = z.object({
+ *   setup: z.string().describe("The setup of the joke"),
+ *   punchline: z.string().describe("The punchline to the joke"),
+ *   rating: z.number().optional().describe("How funny the joke is, from 1 to 10")
+ * }).describe('Joke to tell user.');
  *
- * // Log the response to the console
- * console.log(message);
+ * const structuredLlm = llm.withStructuredOutput(Joke, { name: "Joke" });
+ * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
+ * console.log(jokeResult);
+ * ```
+ *
+ * ```txt
+ * {
+ *   setup: 'Why was the cat sitting on the computer?',
+ *   punchline: 'Because it wanted to keep an eye on the mouse!',
+ *   rating: 7
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>JSON Object Response Format</strong></summary>
+ *
+ * ```typescript
+ * const jsonLlm = llm.bind({ response_format: { type: "json_object" } });
+ * const jsonLlmAiMsg = await jsonLlm.invoke(
+ *   "Return a JSON object with key 'randomInts' and a value of 10 random ints in [0-99]"
+ * );
+ * console.log(jsonLlmAiMsg.content);
+ * ```
+ *
+ * ```txt
+ * {
+ *   "randomInts": [23, 87, 45, 12, 78, 34, 56, 90, 11, 67]
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Multimodal</strong></summary>
+ *
+ * ```typescript
+ * import { HumanMessage } from '@langchain/core/messages';
+ *
+ * const imageUrl = "https://example.com/image.jpg";
+ * const imageData = await fetch(imageUrl).then(res => res.arrayBuffer());
+ * const base64Image = Buffer.from(imageData).toString('base64');
+ *
+ * const message = new HumanMessage({
+ *   content: [
+ *     { type: "text", text: "describe the weather in this image" },
+ *     {
+ *       type: "image_url",
+ *       image_url: { url: `data:image/jpeg;base64,${base64Image}` },
+ *     },
+ *   ]
+ * });
  *
+ * const imageDescriptionAiMsg = await llm.invoke([message]);
+ * console.log(imageDescriptionAiMsg.content);
  * ```
+ *
+ * ```txt
+ * The weather in the image appears to be clear and sunny. The sky is mostly blue with a few scattered white clouds, indicating fair weather. The bright sunlight is casting shadows on the green, grassy hill, suggesting it is a pleasant day with good visibility. There are no signs of rain or stormy conditions.
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Usage Metadata</strong></summary>
+ *
+ * ```typescript
+ * const aiMsgForMetadata = await llm.invoke(input);
+ * console.log(aiMsgForMetadata.usage_metadata);
+ * ```
+ *
+ * ```txt
+ * { input_tokens: 28, output_tokens: 5, total_tokens: 33 }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Logprobs</strong></summary>
+ *
+ * ```typescript
+ * const logprobsLlm = new ChatOpenAI({ logprobs: true });
+ * const aiMsgForLogprobs = await logprobsLlm.invoke(input);
+ * console.log(aiMsgForLogprobs.response_metadata.logprobs);
+ * ```
+ *
+ * ```txt
+ * {
+ *   content: [
+ *     {
+ *       token: 'J',
+ *       logprob: -0.000050616763,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: "'",
+ *       logprob: -0.01868736,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: 'ad',
+ *       logprob: -0.0000030545007,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     { token: 'ore', logprob: 0, bytes: [Array], top_logprobs: [] },
+ *     {
+ *       token: ' la',
+ *       logprob: -0.515404,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     {
+ *       token: ' programm',
+ *       logprob: -0.0000118755715,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     },
+ *     { token: 'ation', logprob: 0, bytes: [Array], top_logprobs: [] },
+ *     {
+ *       token: '.',
+ *       logprob: -0.0000037697225,
+ *       bytes: [Array],
+ *       top_logprobs: []
+ *     }
+ *   ],
+ *   refusal: null
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>Response Metadata</strong></summary>
+ *
+ * ```typescript
+ * const aiMsgForResponseMetadata = await llm.invoke(input);
+ * console.log(aiMsgForResponseMetadata.response_metadata);
+ * ```
+ *
+ * ```txt
+ * {
+ *   tokenUsage: { completionTokens: 5, promptTokens: 28, totalTokens: 33 },
+ *   finish_reason: 'stop',
+ *   system_fingerprint: 'fp_3aa7262c27'
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
+ *
+ * <details>
+ * <summary><strong>JSON Schema Structured Output</strong></summary>
+ *
+ * ```typescript
+ * const llmForJsonSchema = new ChatOpenAI({
+ *   model: "gpt-4o-2024-08-06",
+ * }).withStructuredOutput(
+ *   z.object({
+ *     command: z.string().describe("The command to execute"),
+ *     expectedOutput: z.string().describe("The expected output of the command"),
+ *     options: z
+ *       .array(z.string())
+ *       .describe("The options you can pass to the command"),
+ *   }),
+ *   {
+ *     method: "jsonSchema",
+ *     strict: true, // Optional when using the `jsonSchema` method
+ *   }
+ * );
+ *
+ * const jsonSchemaRes = await llmForJsonSchema.invoke(
+ *   "What is the command to list files in a directory?"
+ * );
+ * console.log(jsonSchemaRes);
+ * ```
+ *
+ * ```txt
+ * {
+ *   command: 'ls',
+ *   expectedOutput: 'A list of files and subdirectories within the specified directory.',
+ *   options: [
+ *     '-a: include directory entries whose names begin with a dot (.).',
+ *     '-l: use a long listing format.',
+ *     '-h: with -l, print sizes in human readable format (e.g., 1K, 234M, 2G).',
+ *     '-t: sort by time, newest first.',
+ *     '-r: reverse order while sorting.',
+ *     '-S: sort by file size, largest first.',
+ *     '-R: list subdirectories recursively.'
+ *   ]
+ * }
+ * ```
+ * </details>
+ *
+ * <br />
  */
 class ChatOpenAI extends chat_models_1.BaseChatModel {
     static lc_name() {
@@ -571,6 +998,17 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             ...kwargs,
         });
     }
+    createResponseFormat(resFormat) {
+        if (resFormat &&
+            resFormat.type === "json_schema" &&
+            resFormat.json_schema.schema &&
+            isZodSchema(resFormat.json_schema.schema)) {
+            return (0, zod_1.zodResponseFormat)(resFormat.json_schema.schema, resFormat.json_schema.name, {
+                description: resFormat.json_schema.description,
+            });
+        }
+        return resFormat;
+    }
     /**
      * Get the parameters used to invoke the model
      */
@@ -610,7 +1048,7 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
                 ? options.tools.map((tool) => _convertChatOpenAIToolTypeToOpenAITool(tool, { strict }))
                 : undefined,
             tool_choice: (0, openai_js_1.formatToOpenAIToolChoice)(options?.tool_choice),
-            response_format: options?.response_format,
+            response_format: this.createResponseFormat(options?.response_format),
             seed: options?.seed,
             ...streamOptionsConfig,
             parallel_tool_calls: options?.parallel_tool_calls,
@@ -636,6 +1074,20 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             stream: true,
         };
         let defaultRole;
+        if (params.response_format &&
+            params.response_format.type === "json_schema") {
+            console.warn(`OpenAI does not yet support streaming with "response_format" set to "json_schema". Falling back to non-streaming mode.`);
+            const res = await this._generate(messages, options, runManager);
+            const chunk = new outputs_1.ChatGenerationChunk({
+                message: new messages_1.AIMessageChunk({
+                    ...res.generations[0].message,
+                }),
+                text: res.generations[0].text,
+                generationInfo: res.generations[0].generationInfo,
+            });
+            yield chunk;
+            return runManager?.handleLLMNewToken(res.generations[0].text ?? "", undefined, undefined, undefined, undefined, { chunk });
+        }
         const streamIterable = await this.completionWithRetry(params, options);
         let usage;
         for await (const data of streamIterable) {
@@ -662,7 +1114,7 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             }
             // eslint-disable-next-line @typescript-eslint/no-explicit-any
             const generationInfo = { ...newTokenIndices };
-            if (choice.finish_reason !== undefined) {
+            if (choice.finish_reason != null) {
                 generationInfo.finish_reason = choice.finish_reason;
                 // Only include system fingerprint in the last chunk for now
                 // to avoid concatenation issues
@@ -739,14 +1191,28 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             return { generations, llmOutput: { estimatedTokenUsage: tokenUsage } };
         }
         else {
-            const data = await this.completionWithRetry({
-                ...params,
-                stream: false,
-                messages: messagesMapped,
-            }, {
-                signal: options?.signal,
-                ...options?.options,
-            });
+            let data;
+            if (options.response_format &&
+                options.response_format.type === "json_schema") {
+                data = await this.betaParsedCompletionWithRetry({
+                    ...params,
+                    stream: false,
+                    messages: messagesMapped,
+                }, {
+                    signal: options?.signal,
+                    ...options?.options,
+                });
+            }
+            else {
+                data = await this.completionWithRetry({
+                    ...params,
+                    stream: false,
+                    messages: messagesMapped,
+                }, {
+                    signal: options?.signal,
+                    ...options?.options,
+                });
+            }
             const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, } = data?.usage ?? {};
             if (completionTokens) {
                 tokenUsage.completionTokens =
@@ -891,6 +1357,25 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             }
         });
     }
+    /**
+     * Call the beta chat completions parse endpoint. This should only be called if
+     * response_format is set to "json_object".
+     * @param {OpenAIClient.Chat.ChatCompletionCreateParamsNonStreaming} request
+     * @param {OpenAICoreRequestOptions | undefined} options
+     */
+    async betaParsedCompletionWithRetry(request, options) {
+        const requestOptions = this._getClientOptions(options);
+        return this.caller.call(async () => {
+            try {
+                const res = await this.client.beta.chat.completions.parse(request, requestOptions);
+                return res;
+            }
+            catch (e) {
+                const error = (0, openai_js_1.wrapOpenAIClientError)(e);
+                throw error;
+            }
+        });
+    }
     _getClientOptions(options) {
         if (!this.client) {
             const openAIEndpointConfig = {
@@ -983,6 +1468,25 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
                 outputParser = new output_parsers_1.JsonOutputParser();
             }
         }
+        else if (method === "jsonSchema") {
+            llm = this.bind({
+                response_format: {
+                    type: "json_schema",
+                    json_schema: {
+                        name: name ?? "extract",
+                        description: schema.description,
+                        schema,
+                        strict: config?.strict,
+                    },
+                },
+            });
+            if (isZodSchema(schema)) {
+                outputParser = output_parsers_1.StructuredOutputParser.fromZodSchema(schema);
+            }
+            else {
+                outputParser = new output_parsers_1.JsonOutputParser();
+            }
+        }
         else {
             let functionName = name ?? "extract";
             // Is function calling