openai[minor]: OpenAI - Ability to set stream_options and streamUsage when instantiating ChatOpenAI class. (#5761)

* fix: add streamOptions and cleaning up

* Added tests. Included streamUsage flag

* Remove unnecessary type

* Moving streamUsage to instantiation

* Moving streamUsage to instantiation in tests

* Altering stream option to be set based on whether streamUsage/include_usage is set to true.

* Keep stream_usage defaulting to true

* Lint

---------

Co-authored-by: jacoblee93 <[email protected]>

Author: deranga

libs/langchain-openai/src/chat_models.ts

@@ -261,6 +261,7 @@ export interface ChatOpenAICallOptions
   seed?: number;
   /**
    * Additional options to pass to streamed completions.
+   * If provided takes precedence over "streamUsage" set at initialization time.
    */
   stream_options?: {
     /**
@@ -385,6 +386,8 @@ export class ChatOpenAI<
 
   streaming = false;
 
+  streamUsage = true;
+
   maxTokens?: number;
 
   logprobs?: boolean;
@@ -481,6 +484,7 @@ export class ChatOpenAI<
     this.user = fields?.user;
 
     this.streaming = fields?.streaming ?? false;
+    this.streamUsage = fields?.streamUsage ?? this.streamUsage;
 
     if (this.azureOpenAIApiKey || this.azureADTokenProvider) {
       if (!this.azureOpenAIApiInstanceName && !this.azureOpenAIBasePath) {
@@ -549,6 +553,12 @@ export class ChatOpenAI<
         )
       );
     }
+    let streamOptionsConfig = {};
+    if (options?.stream_options !== undefined) {
+      streamOptionsConfig = { stream_options: options.stream_options };
+    } else if (this.streamUsage && this.streaming) {
+      streamOptionsConfig = { stream_options: { include_usage: true } };
+    }
     const params: Omit<
       OpenAIClient.Chat.ChatCompletionCreateParams,
       "messages"
@@ -565,6 +575,7 @@ export class ChatOpenAI<
       logit_bias: this.logitBias,
       stop: options?.stop ?? this.stopSequences,
       user: this.user,
+      // if include_usage is set or streamUsage then stream must be set to true.
       stream: this.streaming,
       functions: options?.functions,
       function_call: options?.function_call,
@@ -574,9 +585,7 @@ export class ChatOpenAI<
       tool_choice: options?.tool_choice,
       response_format: options?.response_format,
       seed: options?.seed,
-      ...(options?.stream_options !== undefined
-        ? { stream_options: options.stream_options }
-        : {}),
+      ...streamOptionsConfig,
       parallel_tool_calls: options?.parallel_tool_calls,
       ...this.modelKwargs,
     };

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

@@ -828,6 +828,33 @@ test("streaming: true tokens can be found in usage_metadata field", async () =>
   expect(response?.usage_metadata?.total_tokens).toBeGreaterThan(0);
 });
 
+test("streaming: streamUsage will not override stream_options", async () => {
+  const model = new ChatOpenAI({
+    streaming: true,
+  });
+  const response = await model.invoke("Hello, how are you?", {
+    stream_options: { include_usage: false },
+  });
+  console.log({
+    usage_metadata: response?.usage_metadata,
+  });
+  expect(response).toBeTruthy();
+  expect(response?.usage_metadata).toBeFalsy();
+});
+
+test("streaming: streamUsage default is true", async () => {
+  const model = new ChatOpenAI();
+  const response = await model.invoke("Hello, how are you?");
+  console.log({
+    usage_metadata: response?.usage_metadata,
+  });
+  expect(response).toBeTruthy();
+  expect(response?.usage_metadata).toBeTruthy();
+  expect(response?.usage_metadata?.input_tokens).toBeGreaterThan(0);
+  expect(response?.usage_metadata?.output_tokens).toBeGreaterThan(0);
+  expect(response?.usage_metadata?.total_tokens).toBeGreaterThan(0);
+});
+
 test("populates ID field on AIMessage", async () => {
   const model = new ChatOpenAI();
   const response = await model.invoke("Hell");

libs/langchain-openai/src/types.ts

@@ -38,6 +38,12 @@ export declare interface OpenAIBaseInput {
   /** Whether to stream the results or not. Enabling disables tokenUsage reporting */
   streaming: boolean;
 
+  /**
+   * Whether or not to include token usage data in streamed chunks.
+   * @default true
+   */
+  streamUsage?: boolean;
+
   /**
    * Model name to use
    * Alias for `model`