Add support for disable_streaming, set for o1

langchain-core/src/language_models/chat_models.ts

@@ -74,7 +74,18 @@ export type SerializedLLM = {
 /**
  * Represents the parameters for a base chat model.
  */
-export type BaseChatModelParams = BaseLanguageModelParams;
+export type BaseChatModelParams = BaseLanguageModelParams & {
+  /**
+   * Whether to disable streaming.
+   *
+   * If streaming is bypassed, then `stream()` will defer to
+   * `invoke()`.
+   *
+   * - If true, will always bypass streaming case.
+   * - If false (default), will always use streaming case if available.
+   */
+  disableStreaming?: boolean;
+};
 
 /**
  * Represents the call options for a base chat model.
@@ -152,6 +163,8 @@ export abstract class BaseChatModel<
   // Only ever instantiated in main LangChain
   lc_namespace = ["langchain", "chat_models", this._llmType()];
 
+  disableStreaming = false;
+
   constructor(fields: BaseChatModelParams) {
     super(fields);
   }
@@ -220,7 +233,8 @@ export abstract class BaseChatModel<
     // Subclass check required to avoid double callbacks with default implementation
     if (
       this._streamResponseChunks ===
-      BaseChatModel.prototype._streamResponseChunks
+        BaseChatModel.prototype._streamResponseChunks ||
+      this.disableStreaming
     ) {
       yield this.invoke(input, options);
     } else {

libs/langchain-openai/src/chat_models.ts

@@ -1223,6 +1223,10 @@ export class ChatOpenAI<
       this.streamUsage = false;
     }
 
+    if (this.model === "o1") {
+      this.disableStreaming = true;
+    }
+
     this.streaming = fields?.streaming ?? false;
     this.streamUsage = fields?.streamUsage ?? this.streamUsage;
 

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

@@ -1165,7 +1165,7 @@ describe("Audio output", () => {
   });
 });
 
-test("Can stream o1 requests", async () => {
+test("Can stream o1-mini requests", async () => {
   const model = new ChatOpenAI({
     model: "o1-mini",
   });
@@ -1192,6 +1192,33 @@ test("Can stream o1 requests", async () => {
   expect(numChunks).toBeGreaterThan(3);
 });
 
+test("Doesn't stream o1 requests", async () => {
+  const model = new ChatOpenAI({
+    model: "o1",
+  });
+  const stream = await model.stream(
+    "Write me a very simple hello world program in Python. Ensure it is wrapped in a function called 'hello_world' and has descriptive comments."
+  );
+  let finalMsg: AIMessageChunk | undefined;
+  let numChunks = 0;
+  for await (const chunk of stream) {
+    finalMsg = finalMsg ? concat(finalMsg, chunk) : chunk;
+    numChunks += 1;
+  }
+
+  expect(finalMsg).toBeTruthy();
+  if (!finalMsg) {
+    throw new Error("No final message found");
+  }
+  if (typeof finalMsg.content === "string") {
+    expect(finalMsg.content.length).toBeGreaterThan(10);
+  } else {
+    expect(finalMsg.content.length).toBeGreaterThanOrEqual(1);
+  }
+
+  expect(numChunks).toBe(1);
+});
+
 test("Allows developer messages with o1", async () => {
   const model = new ChatOpenAI({
     model: "o1",