standard-tests[minor]: Improve prompting to force model to call tool

libs/langchain-standard-tests/src/integration_tests/chat_models.ts

@@ -11,6 +11,7 @@
 import { z } from "zod";
 import { StructuredTool } from "@langchain/core/tools";
 import { zodToJsonSchema } from "zod-to-json-schema";
+import { ChatPromptTemplate } from "@langchain/core/prompts";
 import {
   BaseChatModelsTests,
   BaseChatModelsTestsFields,
@@ -37,6 +38,14 @@
   }
 }
 
+const MATH_ADDITION_PROMPT = /* #__PURE__ */ ChatPromptTemplate.fromMessages([
+  [
+    "system",
+    "You are bad at math and must ALWAYS call the {toolName} function.",
+  ],
+  ["human", "What is the sum of 1836281973 and 19973286?"],
+]);
+
 interface ChatModelIntegrationTestsFields<
   CallOptions extends BaseChatModelCallOptions = BaseChatModelCallOptions,
   OutputMessageType extends BaseMessageChunk = BaseMessageChunk,
@@ -228,11 +237,11 @@
       new ToolMessage(functionResult, functionId, functionName),
     ];
 
-    const resultStringContent = await modelWithTools.invoke(
+    const result = await modelWithTools.invoke(
       messagesStringContent,
       callOptions
     );
-    expect(resultStringContent).toBeInstanceOf(this.invokeResponseType);
+    expect(result).toBeInstanceOf(this.invokeResponseType);
   }
 
   /**
@@ -334,11 +343,11 @@
       new HumanMessage("What is 3 + 4"),
     ];
 
-    const resultStringContent = await modelWithTools.invoke(
+    const result = await modelWithTools.invoke(
       messagesStringContent,
       callOptions
     );
-    expect(resultStringContent).toBeInstanceOf(this.invokeResponseType);
+    expect(result).toBeInstanceOf(this.invokeResponseType);
   }
 
   async testWithStructuredOutput() {
@@ -353,13 +362,17 @@
         "withStructuredOutput undefined. Cannot test tool message histories."
       );
     }
-    const modelWithTools = model.withStructuredOutput(adderSchema);
+    const modelWithTools = model.withStructuredOutput(adderSchema, {
+      name: "math_addition",
+    });
 
-    const resultStringContent = await modelWithTools.invoke("What is 1 + 2");
-    expect(resultStringContent.a).toBeDefined();
-    expect([1, 2].includes(resultStringContent.a)).toBeTruthy();
-    expect(resultStringContent.b).toBeDefined();
-    expect([1, 2].includes(resultStringContent.b)).toBeTruthy();
+    const result = await MATH_ADDITION_PROMPT.pipe(modelWithTools).invoke({
+      toolName: "math_addition",
+    });
+    expect(result.a).toBeDefined();
+    expect(typeof result.a).toBe("number");
+    expect(result.b).toBeDefined();
+    expect(typeof result.b).toBe("number");
   }
 
   async testWithStructuredOutputIncludeRaw() {
@@ -376,14 +389,17 @@
     }
     const modelWithTools = model.withStructuredOutput(adderSchema, {
       includeRaw: true,
+      name: "math_addition",
     });
 
-    const resultStringContent = await modelWithTools.invoke("What is 1 + 2");
-    expect(resultStringContent.raw).toBeInstanceOf(this.invokeResponseType);
-    expect(resultStringContent.parsed.a).toBeDefined();
-    expect([1, 2].includes(resultStringContent.parsed.a)).toBeTruthy();
-    expect(resultStringContent.parsed.b).toBeDefined();
-    expect([1, 2].includes(resultStringContent.parsed.b)).toBeTruthy();
+    const result = await MATH_ADDITION_PROMPT.pipe(modelWithTools).invoke({
+      toolName: "math_addition",
+    });
+    expect(result.raw).toBeInstanceOf(this.invokeResponseType);
+    expect(result.parsed.a).toBeDefined();
+    expect(typeof result.parsed.a).toBe("number");
+    expect(result.parsed.b).toBeDefined();
+    expect(typeof result.parsed.b).toBe("number");
   }
 
   async testBindToolsWithOpenAIFormattedTools() {
@@ -409,7 +425,11 @@
       },
     ]);
 
-    const result: AIMessage = await modelWithTools.invoke("What is 1 + 2");
+    const result: AIMessage = await MATH_ADDITION_PROMPT.pipe(
+      modelWithTools
+    ).invoke({
+      toolName: "math_addition",
+    });
     expect(result.tool_calls).toHaveLength(1);
     if (!result.tool_calls) {
       throw new Error("result.tool_calls is undefined");
@@ -429,49 +449,49 @@
 
     try {
       await this.testInvoke();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testInvoke failed", e);
     }
 
     try {
       await this.testStream();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testStream failed", e);
     }
 
     try {
       await this.testBatch();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testBatch failed", e);
     }
 
     try {
       await this.testConversation();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testConversation failed", e);
     }
 
     try {
       await this.testUsageMetadata();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testUsageMetadata failed", e);
     }
 
     try {
       await this.testUsageMetadataStreaming();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testUsageMetadataStreaming failed", e);
     }
 
     try {
       await this.testToolMessageHistoriesStringContent();
     } catch (e: any) {
       allTestsPassed = false;
       console.error("testToolMessageHistoriesStringContent failed", e);
     }