@langchain/core 0.3.4 → 0.3.5

package/dist/language_models/chat_models.cjs

@@ -88,6 +88,8 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
             };
             const runManagers = await callbackManager_?.handleChatModelStart(this.toJSON(), [messages], runnableConfig.runId, undefined, extra, undefined, undefined, runnableConfig.runName);
             let generationChunk;
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            let llmOutput;
             try {
                 for await (const chunk of this._streamResponseChunks(messages, callOptions, runManagers?.[0])) {
                     if (chunk.message.id == null) {
@@ -106,6 +108,16 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
                     else {
                         generationChunk = generationChunk.concat(chunk);
                     }
+                    if ((0, index_js_1.isAIMessageChunk)(chunk.message) &&
+                        chunk.message.usage_metadata !== undefined) {
+                        llmOutput = {
+                            tokenUsage: {
+                                promptTokens: chunk.message.usage_metadata.input_tokens,
+                                completionTokens: chunk.message.usage_metadata.output_tokens,
+                                totalTokens: chunk.message.usage_metadata.total_tokens,
+                            },
+                        };
+                    }
                 }
             }
             catch (err) {
@@ -115,6 +127,7 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
             await Promise.all((runManagers ?? []).map((runManager) => runManager?.handleLLMEnd({
                 // TODO: Remove cast after figuring out inheritance
                 generations: [[generationChunk]],
+                llmOutput,
             })));
         }
     }
@@ -154,6 +167,8 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
             try {
                 const stream = await this._streamResponseChunks(baseMessages[0], parsedOptions, runManagers?.[0]);
                 let aggregated;
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let llmOutput;
                 for await (const chunk of stream) {
                     if (chunk.message.id == null) {
                         const runId = runManagers?.at(0)?.runId;
@@ -166,6 +181,16 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
                     else {
                         aggregated = (0, stream_js_1.concat)(aggregated, chunk);
                     }
+                    if ((0, index_js_1.isAIMessageChunk)(chunk.message) &&
+                        chunk.message.usage_metadata !== undefined) {
+                        llmOutput = {
+                            tokenUsage: {
+                                promptTokens: chunk.message.usage_metadata.input_tokens,
+                                completionTokens: chunk.message.usage_metadata.output_tokens,
+                                totalTokens: chunk.message.usage_metadata.total_tokens,
+                            },
+                        };
+                    }
                 }
                 if (aggregated === undefined) {
                     throw new Error("Received empty response from chat model call.");
@@ -173,7 +198,7 @@ class BaseChatModel extends base_js_1.BaseLanguageModel {
                 generations.push([aggregated]);
                 await runManagers?.[0].handleLLMEnd({
                     generations,
-                    llmOutput: {},
+                    llmOutput,
                 });
             }
             catch (e) {

package/dist/language_models/chat_models.d.ts

@@ -1,5 +1,5 @@
 import { z } from "zod";
-import { type BaseMessage, BaseMessageChunk, type BaseMessageLike } from "../messages/index.js";
+import { type BaseMessage, BaseMessageChunk, type BaseMessageLike, AIMessageChunk } from "../messages/index.js";
 import type { BasePromptValueInterface } from "../prompt_values.js";
 import { LLMResult, ChatGenerationChunk, type ChatResult, type Generation } from "../outputs.js";
 import { BaseLanguageModel, type StructuredOutputMethodOptions, type ToolDefinition, type BaseLanguageModelCallOptions, type BaseLanguageModelInput, type BaseLanguageModelParams } from "./base.js";
@@ -66,7 +66,7 @@ export type BindToolsInput = StructuredToolInterface | Record<string, any> | Too
  * Base class for chat models. It extends the BaseLanguageModel class and
  * provides methods for generating chat based on input messages.
  */
-export declare abstract class BaseChatModel<CallOptions extends BaseChatModelCallOptions = BaseChatModelCallOptions, OutputMessageType extends BaseMessageChunk = BaseMessageChunk> extends BaseLanguageModel<OutputMessageType, CallOptions> {
+export declare abstract class BaseChatModel<CallOptions extends BaseChatModelCallOptions = BaseChatModelCallOptions, OutputMessageType extends BaseMessageChunk = AIMessageChunk> extends BaseLanguageModel<OutputMessageType, CallOptions> {
     ParsedCallOptions: Omit<CallOptions, Exclude<keyof RunnableConfig, "signal" | "timeout" | "maxConcurrency">>;
     lc_namespace: string[];
     constructor(fields: BaseChatModelParams);

package/dist/language_models/chat_models.js

@@ -1,5 +1,5 @@
 import { zodToJsonSchema } from "zod-to-json-schema";
-import { AIMessage, HumanMessage, coerceMessageLikeToMessage, } from "../messages/index.js";
+import { AIMessage, HumanMessage, coerceMessageLikeToMessage, isAIMessageChunk, } from "../messages/index.js";
 import { RUN_KEY, } from "../outputs.js";
 import { BaseLanguageModel, } from "./base.js";
 import { CallbackManager, } from "../callbacks/manager.js";
@@ -84,6 +84,8 @@ export class BaseChatModel extends BaseLanguageModel {
             };
             const runManagers = await callbackManager_?.handleChatModelStart(this.toJSON(), [messages], runnableConfig.runId, undefined, extra, undefined, undefined, runnableConfig.runName);
             let generationChunk;
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            let llmOutput;
             try {
                 for await (const chunk of this._streamResponseChunks(messages, callOptions, runManagers?.[0])) {
                     if (chunk.message.id == null) {
@@ -102,6 +104,16 @@ export class BaseChatModel extends BaseLanguageModel {
                     else {
                         generationChunk = generationChunk.concat(chunk);
                     }
+                    if (isAIMessageChunk(chunk.message) &&
+                        chunk.message.usage_metadata !== undefined) {
+                        llmOutput = {
+                            tokenUsage: {
+                                promptTokens: chunk.message.usage_metadata.input_tokens,
+                                completionTokens: chunk.message.usage_metadata.output_tokens,
+                                totalTokens: chunk.message.usage_metadata.total_tokens,
+                            },
+                        };
+                    }
                 }
             }
             catch (err) {
@@ -111,6 +123,7 @@ export class BaseChatModel extends BaseLanguageModel {
             await Promise.all((runManagers ?? []).map((runManager) => runManager?.handleLLMEnd({
                 // TODO: Remove cast after figuring out inheritance
                 generations: [[generationChunk]],
+                llmOutput,
             })));
         }
     }
@@ -150,6 +163,8 @@ export class BaseChatModel extends BaseLanguageModel {
             try {
                 const stream = await this._streamResponseChunks(baseMessages[0], parsedOptions, runManagers?.[0]);
                 let aggregated;
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let llmOutput;
                 for await (const chunk of stream) {
                     if (chunk.message.id == null) {
                         const runId = runManagers?.at(0)?.runId;
@@ -162,6 +177,16 @@ export class BaseChatModel extends BaseLanguageModel {
                     else {
                         aggregated = concat(aggregated, chunk);
                     }
+                    if (isAIMessageChunk(chunk.message) &&
+                        chunk.message.usage_metadata !== undefined) {
+                        llmOutput = {
+                            tokenUsage: {
+                                promptTokens: chunk.message.usage_metadata.input_tokens,
+                                completionTokens: chunk.message.usage_metadata.output_tokens,
+                                totalTokens: chunk.message.usage_metadata.total_tokens,
+                            },
+                        };
+                    }
                 }
                 if (aggregated === undefined) {
                     throw new Error("Received empty response from chat model call.");
@@ -169,7 +194,7 @@ export class BaseChatModel extends BaseLanguageModel {
                 generations.push([aggregated]);
                 await runManagers?.[0].handleLLMEnd({
                     generations,
-                    llmOutput: {},
+                    llmOutput,
                 });
             }
             catch (e) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@langchain/core",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "description": "Core LangChain.js abstractions and schemas",
   "type": "module",
   "engines": {