@langchain/openai 0.3.3 → 0.3.5

package/dist/chat_models.cjs

@@ -74,6 +74,7 @@ function openAIResponseToChatMessage(message, rawResponse, includeRawResponse) {
             let response_metadata;
             if (rawResponse.system_fingerprint) {
                 response_metadata = {
+                    usage: { ...rawResponse.usage },
                     system_fingerprint: rawResponse.system_fingerprint,
                 };
             }
@@ -112,8 +113,9 @@ delta, rawResponse, defaultRole, includeRawResponse) {
     if (includeRawResponse) {
         additional_kwargs.__raw_response = rawResponse;
     }
+    const response_metadata = { usage: { ...rawResponse.usage } };
     if (role === "user") {
-        return new messages_1.HumanMessageChunk({ content });
+        return new messages_1.HumanMessageChunk({ content, response_metadata });
     }
     else if (role === "assistant") {
         const toolCallChunks = [];
@@ -133,16 +135,18 @@ delta, rawResponse, defaultRole, includeRawResponse) {
             tool_call_chunks: toolCallChunks,
             additional_kwargs,
             id: rawResponse.id,
+            response_metadata,
         });
     }
     else if (role === "system") {
-        return new messages_1.SystemMessageChunk({ content });
+        return new messages_1.SystemMessageChunk({ content, response_metadata });
     }
     else if (role === "function") {
         return new messages_1.FunctionMessageChunk({
             content,
             additional_kwargs,
             name: delta.name,
+            response_metadata,
         });
     }
     else if (role === "tool") {
@@ -150,10 +154,11 @@ delta, rawResponse, defaultRole, includeRawResponse) {
             content,
             additional_kwargs,
             tool_call_id: delta.tool_call_id,
+            response_metadata,
         });
     }
     else {
-        return new messages_1.ChatMessageChunk({ content, role });
+        return new messages_1.ChatMessageChunk({ content, role, response_metadata });
     }
 }
 // Used in LangSmith, export is important here
@@ -458,7 +463,10 @@ function _convertChatOpenAIToolTypeToOpenAITool(tool, fields) {
  *   rating: z.number().optional().describe("How funny the joke is, from 1 to 10")
  * }).describe('Joke to tell user.');
  *
- * const structuredLlm = llm.withStructuredOutput(Joke, { name: "Joke" });
+ * const structuredLlm = llm.withStructuredOutput(Joke, {
+ *   name: "Joke",
+ *   strict: true, // Optionally enable OpenAI structured outputs
+ * });
  * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
  * console.log(jokeResult);
  * ```
@@ -1105,20 +1113,6 @@ 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) {
@@ -1166,6 +1160,9 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             const generationChunk = new outputs_1.ChatGenerationChunk({
                 message: new messages_1.AIMessageChunk({
                     content: "",
+                    response_metadata: {
+                        usage: { ...usage },
+                    },
                     usage_metadata: {
                         input_tokens: usage.prompt_tokens,
                         output_tokens: usage.completion_tokens,

package/dist/chat_models.d.ts

@@ -334,7 +334,10 @@ export interface ChatOpenAIFields extends Partial<OpenAIChatInput>, Partial<Azur
  *   rating: z.number().optional().describe("How funny the joke is, from 1 to 10")
  * }).describe('Joke to tell user.');
  *
- * const structuredLlm = llm.withStructuredOutput(Joke, { name: "Joke" });
+ * const structuredLlm = llm.withStructuredOutput(Joke, {
+ *   name: "Joke",
+ *   strict: true, // Optionally enable OpenAI structured outputs
+ * });
  * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
  * console.log(jokeResult);
  * ```

package/dist/chat_models.js

@@ -70,6 +70,7 @@ function openAIResponseToChatMessage(message, rawResponse, includeRawResponse) {
             let response_metadata;
             if (rawResponse.system_fingerprint) {
                 response_metadata = {
+                    usage: { ...rawResponse.usage },
                     system_fingerprint: rawResponse.system_fingerprint,
                 };
             }
@@ -108,8 +109,9 @@ delta, rawResponse, defaultRole, includeRawResponse) {
     if (includeRawResponse) {
         additional_kwargs.__raw_response = rawResponse;
     }
+    const response_metadata = { usage: { ...rawResponse.usage } };
     if (role === "user") {
-        return new HumanMessageChunk({ content });
+        return new HumanMessageChunk({ content, response_metadata });
     }
     else if (role === "assistant") {
         const toolCallChunks = [];
@@ -129,16 +131,18 @@ delta, rawResponse, defaultRole, includeRawResponse) {
             tool_call_chunks: toolCallChunks,
             additional_kwargs,
             id: rawResponse.id,
+            response_metadata,
         });
     }
     else if (role === "system") {
-        return new SystemMessageChunk({ content });
+        return new SystemMessageChunk({ content, response_metadata });
     }
     else if (role === "function") {
         return new FunctionMessageChunk({
             content,
             additional_kwargs,
             name: delta.name,
+            response_metadata,
         });
     }
     else if (role === "tool") {
@@ -146,10 +150,11 @@ delta, rawResponse, defaultRole, includeRawResponse) {
             content,
             additional_kwargs,
             tool_call_id: delta.tool_call_id,
+            response_metadata,
         });
     }
     else {
-        return new ChatMessageChunk({ content, role });
+        return new ChatMessageChunk({ content, role, response_metadata });
     }
 }
 // Used in LangSmith, export is important here
@@ -453,7 +458,10 @@ function _convertChatOpenAIToolTypeToOpenAITool(tool, fields) {
  *   rating: z.number().optional().describe("How funny the joke is, from 1 to 10")
  * }).describe('Joke to tell user.');
  *
- * const structuredLlm = llm.withStructuredOutput(Joke, { name: "Joke" });
+ * const structuredLlm = llm.withStructuredOutput(Joke, {
+ *   name: "Joke",
+ *   strict: true, // Optionally enable OpenAI structured outputs
+ * });
  * const jokeResult = await structuredLlm.invoke("Tell me a joke about cats");
  * console.log(jokeResult);
  * ```
@@ -1100,20 +1108,6 @@ export class ChatOpenAI extends 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 ChatGenerationChunk({
-                message: new 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) {
@@ -1161,6 +1155,9 @@ export class ChatOpenAI extends BaseChatModel {
             const generationChunk = new ChatGenerationChunk({
                 message: new AIMessageChunk({
                     content: "",
+                    response_metadata: {
+                        usage: { ...usage },
+                    },
                     usage_metadata: {
                         input_tokens: usage.prompt_tokens,
                         output_tokens: usage.completion_tokens,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@langchain/openai",
-  "version": "0.3.3",
+  "version": "0.3.5",
   "description": "OpenAI integrations for LangChain.js",
   "type": "module",
   "engines": {