core: Add signal/timeout options to RunnableConfig

langchain-core/src/language_models/base.ts

@@ -207,18 +207,6 @@ export interface BaseLanguageModelCallOptions extends RunnableConfig {
    * If not provided, the default stop tokens for the model will be used.
    */
   stop?: string[];
-
-  /**
-   * Timeout for this call in milliseconds.
-   */
-  timeout?: number;
-
-  /**
-   * Abort signal for this call.
-   * If provided, the call will be aborted when the signal is aborted.
-   * @see https://developer.mozilla.org/en-US/docs/Web/API/AbortSignal
-   */
-  signal?: AbortSignal;
 }
 
 export interface FunctionDefinition {

langchain-core/src/language_models/chat_models.ts

@@ -145,7 +145,7 @@ export abstract class BaseChatModel<
 > extends BaseLanguageModel<OutputMessageType, CallOptions> {
   declare ParsedCallOptions: Omit<
     CallOptions,
-    keyof RunnableConfig & "timeout"
+    Exclude<keyof RunnableConfig, "signal">
   >;
 
   // Only ever instantiated in main LangChain
@@ -159,14 +159,13 @@ export abstract class BaseChatModel<
     ...llmOutputs: LLMResult["llmOutput"][]
   ): LLMResult["llmOutput"];
 
-  protected _separateRunnableConfigFromCallOptions(
+  protected _separateRunnableConfigFromCallOptionsCompat(
     options?: Partial<CallOptions>
   ): [RunnableConfig, this["ParsedCallOptions"]] {
+    // For backwards compat, keep `signal` in both runnableConfig and callOptions
     const [runnableConfig, callOptions] =
       super._separateRunnableConfigFromCallOptions(options);
-    if (callOptions?.timeout && !callOptions.signal) {
-      callOptions.signal = AbortSignal.timeout(callOptions.timeout);
-    }
+    (callOptions as this["ParsedCallOptions"]).signal = runnableConfig.signal;
     return [runnableConfig, callOptions as this["ParsedCallOptions"]];
   }
 
@@ -232,7 +231,7 @@ export abstract class BaseChatModel<
       const prompt = BaseChatModel._convertInputToPromptValue(input);
       const messages = prompt.toChatMessages();
       const [runnableConfig, callOptions] =
-        this._separateRunnableConfigFromCallOptions(options);
+        this._separateRunnableConfigFromCallOptionsCompat(options);
 
       const inheritableMetadata = {
         ...runnableConfig.metadata,
@@ -578,16 +577,17 @@ export abstract class BaseChatModel<
     );
 
     const [runnableConfig, callOptions] =
-      this._separateRunnableConfigFromCallOptions(parsedOptions);
+      this._separateRunnableConfigFromCallOptionsCompat(parsedOptions);
     runnableConfig.callbacks = runnableConfig.callbacks ?? callbacks;
 
     if (!this.cache) {
       return this._generateUncached(baseMessages, callOptions, runnableConfig);
     }
 
     const { cache } = this;
-    const llmStringKey =
-      this._getSerializedCacheKeyParametersForCall(callOptions);
+    const llmStringKey = this._getSerializedCacheKeyParametersForCall(
+      callOptions as CallOptions
+    );
 
     const { generations, missingPromptIndices } = await this._generateCached({
       messages: baseMessages,

langchain-core/src/language_models/llms.ts

@@ -63,7 +63,7 @@ export abstract class BaseLLM<
 > extends BaseLanguageModel<string, CallOptions> {
   declare ParsedCallOptions: Omit<
     CallOptions,
-    keyof RunnableConfig & "timeout"
+    Exclude<keyof RunnableConfig, "signal">
   >;
 
   // Only ever instantiated in main LangChain
@@ -103,14 +103,13 @@ export abstract class BaseLLM<
     throw new Error("Not implemented.");
   }
 
-  protected _separateRunnableConfigFromCallOptions(
+  protected _separateRunnableConfigFromCallOptionsCompat(
     options?: Partial<CallOptions>
   ): [RunnableConfig, this["ParsedCallOptions"]] {
+    // For backwards compat, keep `signal` in both runnableConfig and callOptions
     const [runnableConfig, callOptions] =
       super._separateRunnableConfigFromCallOptions(options);
-    if (callOptions?.timeout && !callOptions.signal) {
-      callOptions.signal = AbortSignal.timeout(callOptions.timeout);
-    }
+    (callOptions as this["ParsedCallOptions"]).signal = runnableConfig.signal;
     return [runnableConfig, callOptions as this["ParsedCallOptions"]];
   }
 
@@ -126,7 +125,7 @@ export abstract class BaseLLM<
     } else {
       const prompt = BaseLLM._convertInputToPromptValue(input);
       const [runnableConfig, callOptions] =
-        this._separateRunnableConfigFromCallOptions(options);
+        this._separateRunnableConfigFromCallOptionsCompat(options);
       const callbackManager_ = await CallbackManager.configure(
         runnableConfig.callbacks,
         this.callbacks,
@@ -461,16 +460,17 @@ export abstract class BaseLLM<
     }
 
     const [runnableConfig, callOptions] =
-      this._separateRunnableConfigFromCallOptions(parsedOptions);
+      this._separateRunnableConfigFromCallOptionsCompat(parsedOptions);
     runnableConfig.callbacks = runnableConfig.callbacks ?? callbacks;
 
     if (!this.cache) {
       return this._generateUncached(prompts, callOptions, runnableConfig);
     }
 
     const { cache } = this;
-    const llmStringKey =
-      this._getSerializedCacheKeyParametersForCall(callOptions);
+    const llmStringKey = this._getSerializedCacheKeyParametersForCall(
+      callOptions as CallOptions
+    );
     const { generations, missingPromptIndices } = await this._generateCached({
       prompts,
       cache,