@langchain/openai 0.3.4 → 0.3.6-rc.0

package/dist/azure/embeddings.d.ts

@@ -10,5 +10,7 @@ export declare class AzureOpenAIEmbeddings extends OpenAIEmbeddings {
         deploymentName?: string;
         openAIApiVersion?: string;
     }, configuration?: ClientOptions & LegacyOpenAIInput);
-    protected embeddingWithRetry(request: OpenAIClient.EmbeddingCreateParams): Promise<OpenAIClient.Embeddings.CreateEmbeddingResponse>;
+    protected embeddingWithRetry(request: OpenAIClient.EmbeddingCreateParams): Promise<OpenAIClient.Embeddings.CreateEmbeddingResponse & {
+        _request_id?: string | null | undefined;
+    }>;
 }

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
@@ -1152,13 +1157,38 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             await runManager?.handleLLMNewToken(generationChunk.text ?? "", newTokenIndices, undefined, undefined, undefined, { chunk: generationChunk });
         }
         if (usage) {
+            const inputTokenDetails = {
+                ...(usage.prompt_tokens_details?.audio_tokens !== null && {
+                    audio: usage.prompt_tokens_details?.audio_tokens,
+                }),
+                ...(usage.prompt_tokens_details?.cached_tokens !== null && {
+                    cache_read: usage.prompt_tokens_details?.cached_tokens,
+                }),
+            };
+            const outputTokenDetails = {
+                ...(usage.completion_tokens_details?.audio_tokens !== null && {
+                    audio: usage.completion_tokens_details?.audio_tokens,
+                }),
+                ...(usage.completion_tokens_details?.reasoning_tokens !== null && {
+                    reasoning: usage.completion_tokens_details?.reasoning_tokens,
+                }),
+            };
             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,
                         total_tokens: usage.total_tokens,
+                        ...(Object.keys(inputTokenDetails).length > 0 && {
+                            input_token_details: inputTokenDetails,
+                        }),
+                        ...(Object.keys(outputTokenDetails).length > 0 && {
+                            output_token_details: outputTokenDetails,
+                        }),
                     },
                 }),
                 text: "",
@@ -1178,7 +1208,7 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
     }
     /** @ignore */
     async _generate(messages, options, runManager) {
-        const tokenUsage = {};
+        const usageMetadata = {};
         const params = this.invocationParams(options);
         const messagesMapped = _convertMessagesToOpenAIParams(messages);
         if (params.stream) {
@@ -1205,10 +1235,19 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
             // fallback to estimation.
             const promptTokenUsage = await this.getEstimatedTokenCountFromPrompt(messages, functions, function_call);
             const completionTokenUsage = await this.getNumTokensFromGenerations(generations);
-            tokenUsage.promptTokens = promptTokenUsage;
-            tokenUsage.completionTokens = completionTokenUsage;
-            tokenUsage.totalTokens = promptTokenUsage + completionTokenUsage;
-            return { generations, llmOutput: { estimatedTokenUsage: tokenUsage } };
+            usageMetadata.input_tokens = promptTokenUsage;
+            usageMetadata.output_tokens = completionTokenUsage;
+            usageMetadata.total_tokens = promptTokenUsage + completionTokenUsage;
+            return {
+                generations,
+                llmOutput: {
+                    estimatedTokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
+                },
+            };
         }
         else {
             let data;
@@ -1233,16 +1272,40 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
                     ...options?.options,
                 });
             }
-            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, } = data?.usage ?? {};
+            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, prompt_tokens_details: promptTokensDetails, completion_tokens_details: completionTokensDetails, } = data?.usage ?? {};
             if (completionTokens) {
-                tokenUsage.completionTokens =
-                    (tokenUsage.completionTokens ?? 0) + completionTokens;
+                usageMetadata.output_tokens =
+                    (usageMetadata.output_tokens ?? 0) + completionTokens;
             }
             if (promptTokens) {
-                tokenUsage.promptTokens = (tokenUsage.promptTokens ?? 0) + promptTokens;
+                usageMetadata.input_tokens =
+                    (usageMetadata.input_tokens ?? 0) + promptTokens;
             }
             if (totalTokens) {
-                tokenUsage.totalTokens = (tokenUsage.totalTokens ?? 0) + totalTokens;
+                usageMetadata.total_tokens =
+                    (usageMetadata.total_tokens ?? 0) + totalTokens;
+            }
+            if (promptTokensDetails?.audio_tokens !== null ||
+                promptTokensDetails?.cached_tokens !== null) {
+                usageMetadata.input_token_details = {
+                    ...(promptTokensDetails?.audio_tokens !== null && {
+                        audio: promptTokensDetails?.audio_tokens,
+                    }),
+                    ...(promptTokensDetails?.cached_tokens !== null && {
+                        cache_read: promptTokensDetails?.cached_tokens,
+                    }),
+                };
+            }
+            if (completionTokensDetails?.audio_tokens !== null ||
+                completionTokensDetails?.reasoning_tokens !== null) {
+                usageMetadata.output_token_details = {
+                    ...(completionTokensDetails?.audio_tokens !== null && {
+                        audio: completionTokensDetails?.audio_tokens,
+                    }),
+                    ...(completionTokensDetails?.reasoning_tokens !== null && {
+                        reasoning: completionTokensDetails?.reasoning_tokens,
+                    }),
+                };
             }
             const generations = [];
             for (const part of data?.choices ?? []) {
@@ -1256,17 +1319,19 @@ class ChatOpenAI extends chat_models_1.BaseChatModel {
                     ...(part.logprobs ? { logprobs: part.logprobs } : {}),
                 };
                 if ((0, messages_1.isAIMessage)(generation.message)) {
-                    generation.message.usage_metadata = {
-                        input_tokens: tokenUsage.promptTokens ?? 0,
-                        output_tokens: tokenUsage.completionTokens ?? 0,
-                        total_tokens: tokenUsage.totalTokens ?? 0,
-                    };
+                    generation.message.usage_metadata = usageMetadata;
                 }
                 generations.push(generation);
             }
             return {
                 generations,
-                llmOutput: { tokenUsage },
+                llmOutput: {
+                    tokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
+                },
             };
         }
     }

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
@@ -1147,13 +1152,38 @@ export class ChatOpenAI extends BaseChatModel {
             await runManager?.handleLLMNewToken(generationChunk.text ?? "", newTokenIndices, undefined, undefined, undefined, { chunk: generationChunk });
         }
         if (usage) {
+            const inputTokenDetails = {
+                ...(usage.prompt_tokens_details?.audio_tokens !== null && {
+                    audio: usage.prompt_tokens_details?.audio_tokens,
+                }),
+                ...(usage.prompt_tokens_details?.cached_tokens !== null && {
+                    cache_read: usage.prompt_tokens_details?.cached_tokens,
+                }),
+            };
+            const outputTokenDetails = {
+                ...(usage.completion_tokens_details?.audio_tokens !== null && {
+                    audio: usage.completion_tokens_details?.audio_tokens,
+                }),
+                ...(usage.completion_tokens_details?.reasoning_tokens !== null && {
+                    reasoning: usage.completion_tokens_details?.reasoning_tokens,
+                }),
+            };
             const generationChunk = new ChatGenerationChunk({
                 message: new AIMessageChunk({
                     content: "",
+                    response_metadata: {
+                        usage: { ...usage },
+                    },
                     usage_metadata: {
                         input_tokens: usage.prompt_tokens,
                         output_tokens: usage.completion_tokens,
                         total_tokens: usage.total_tokens,
+                        ...(Object.keys(inputTokenDetails).length > 0 && {
+                            input_token_details: inputTokenDetails,
+                        }),
+                        ...(Object.keys(outputTokenDetails).length > 0 && {
+                            output_token_details: outputTokenDetails,
+                        }),
                     },
                 }),
                 text: "",
@@ -1173,7 +1203,7 @@ export class ChatOpenAI extends BaseChatModel {
     }
     /** @ignore */
     async _generate(messages, options, runManager) {
-        const tokenUsage = {};
+        const usageMetadata = {};
         const params = this.invocationParams(options);
         const messagesMapped = _convertMessagesToOpenAIParams(messages);
         if (params.stream) {
@@ -1200,10 +1230,19 @@ export class ChatOpenAI extends BaseChatModel {
             // fallback to estimation.
             const promptTokenUsage = await this.getEstimatedTokenCountFromPrompt(messages, functions, function_call);
             const completionTokenUsage = await this.getNumTokensFromGenerations(generations);
-            tokenUsage.promptTokens = promptTokenUsage;
-            tokenUsage.completionTokens = completionTokenUsage;
-            tokenUsage.totalTokens = promptTokenUsage + completionTokenUsage;
-            return { generations, llmOutput: { estimatedTokenUsage: tokenUsage } };
+            usageMetadata.input_tokens = promptTokenUsage;
+            usageMetadata.output_tokens = completionTokenUsage;
+            usageMetadata.total_tokens = promptTokenUsage + completionTokenUsage;
+            return {
+                generations,
+                llmOutput: {
+                    estimatedTokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
+                },
+            };
         }
         else {
             let data;
@@ -1228,16 +1267,40 @@ export class ChatOpenAI extends BaseChatModel {
                     ...options?.options,
                 });
             }
-            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, } = data?.usage ?? {};
+            const { completion_tokens: completionTokens, prompt_tokens: promptTokens, total_tokens: totalTokens, prompt_tokens_details: promptTokensDetails, completion_tokens_details: completionTokensDetails, } = data?.usage ?? {};
             if (completionTokens) {
-                tokenUsage.completionTokens =
-                    (tokenUsage.completionTokens ?? 0) + completionTokens;
+                usageMetadata.output_tokens =
+                    (usageMetadata.output_tokens ?? 0) + completionTokens;
             }
             if (promptTokens) {
-                tokenUsage.promptTokens = (tokenUsage.promptTokens ?? 0) + promptTokens;
+                usageMetadata.input_tokens =
+                    (usageMetadata.input_tokens ?? 0) + promptTokens;
             }
             if (totalTokens) {
-                tokenUsage.totalTokens = (tokenUsage.totalTokens ?? 0) + totalTokens;
+                usageMetadata.total_tokens =
+                    (usageMetadata.total_tokens ?? 0) + totalTokens;
+            }
+            if (promptTokensDetails?.audio_tokens !== null ||
+                promptTokensDetails?.cached_tokens !== null) {
+                usageMetadata.input_token_details = {
+                    ...(promptTokensDetails?.audio_tokens !== null && {
+                        audio: promptTokensDetails?.audio_tokens,
+                    }),
+                    ...(promptTokensDetails?.cached_tokens !== null && {
+                        cache_read: promptTokensDetails?.cached_tokens,
+                    }),
+                };
+            }
+            if (completionTokensDetails?.audio_tokens !== null ||
+                completionTokensDetails?.reasoning_tokens !== null) {
+                usageMetadata.output_token_details = {
+                    ...(completionTokensDetails?.audio_tokens !== null && {
+                        audio: completionTokensDetails?.audio_tokens,
+                    }),
+                    ...(completionTokensDetails?.reasoning_tokens !== null && {
+                        reasoning: completionTokensDetails?.reasoning_tokens,
+                    }),
+                };
             }
             const generations = [];
             for (const part of data?.choices ?? []) {
@@ -1251,17 +1314,19 @@ export class ChatOpenAI extends BaseChatModel {
                     ...(part.logprobs ? { logprobs: part.logprobs } : {}),
                 };
                 if (isAIMessage(generation.message)) {
-                    generation.message.usage_metadata = {
-                        input_tokens: tokenUsage.promptTokens ?? 0,
-                        output_tokens: tokenUsage.completionTokens ?? 0,
-                        total_tokens: tokenUsage.totalTokens ?? 0,
-                    };
+                    generation.message.usage_metadata = usageMetadata;
                 }
                 generations.push(generation);
             }
             return {
                 generations,
-                llmOutput: { tokenUsage },
+                llmOutput: {
+                    tokenUsage: {
+                        promptTokens: usageMetadata.input_tokens,
+                        completionTokens: usageMetadata.output_tokens,
+                        totalTokens: usageMetadata.total_tokens,
+                    },
+                },
             };
         }
     }

package/dist/embeddings.d.ts

@@ -102,5 +102,10 @@ export declare class OpenAIEmbeddings extends Embeddings implements OpenAIEmbedd
      * @param request Request to send to the OpenAI API.
      * @returns Promise that resolves to the response from the API.
      */
-    protected embeddingWithRetry(request: OpenAIClient.EmbeddingCreateParams): Promise<OpenAIClient.Embeddings.CreateEmbeddingResponse>;
+    protected embeddingWithRetry(request: OpenAIClient.EmbeddingCreateParams): Promise<OpenAIClient.Embeddings.CreateEmbeddingResponse & {
+        _request_id?: string | null | undefined; /**
+         * The number of dimensions the resulting output embeddings should have.
+         * Only supported in `text-embedding-3` and later models.
+         */
+    }>;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@langchain/openai",
-  "version": "0.3.4",
+  "version": "0.3.6-rc.0",
   "description": "OpenAI integrations for LangChain.js",
   "type": "module",
   "engines": {
@@ -36,7 +36,7 @@
   "license": "MIT",
   "dependencies": {
     "js-tiktoken": "^1.0.12",
-    "openai": "^4.57.3",
+    "openai": "^4.67.2",
     "zod": "^3.22.4",
     "zod-to-json-schema": "^3.22.3"
   },