perf(intent): doubao-seed-2.0 分类请求关闭深度思考

src/main/java/com/wok/supportbot/service/IntentRouter.java

@@ -8,6 +8,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.ai.chat.model.ChatModel;
 import org.springframework.ai.chat.prompt.Prompt;
 import org.springframework.ai.converter.BeanOutputConverter;
+import org.springframework.ai.openai.OpenAiChatOptions;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Service;
 
@@ -33,6 +34,9 @@ public class IntentRouter {
     private static final BeanOutputConverter<IntentResult> INTENT_CONVERTER =
             new BeanOutputConverter<>(IntentResult.class);
 
+    /** 分类只需返回 intent/confidence；不改变主回答模型的深度思考与输出长度。 */
+    private static final int CLASSIFICATION_MAX_TOKENS = 128;
+
     /**
      * 意图分类 Prompt 模板。
      * 输出格式约束（JSON Schema）由 {@code BeanOutputConverter.getFormat()} 统一追加，模板中不再硬编码。
@@ -80,7 +84,8 @@ public class IntentRouter {
             ChatModel chatModel = chatModelFactory.getChatModel("CHAT");
             String promptText = INTENT_PROMPT_TEMPLATE.formatted(userQuestion, INTENT_CONVERTER.getFormat());
 
-            String response = chatModel.call(new Prompt(promptText)).getResult().getOutput().getText();
+            String response = chatModel.call(classificationPrompt(chatModel, promptText))
+                    .getResult().getOutput().getText();
             log.debug("意图分类原始响应: {}", response);
 
             IntentResult result = INTENT_CONVERTER.convert(response);
@@ -95,6 +100,21 @@ public class IntentRouter {
         }
     }
 
+    private Prompt classificationPrompt(ChatModel chatModel, String promptText) {
+        // Seed 2.0 默认开启深度思考；官方将 minimal 映射为关闭思考。
+        // 仅为已支持的模型系列覆盖本次分类请求，其他提供商不发送此参数。
+        // https://www.volcengine.com/docs/82379/1449737
+        if (chatModel.getDefaultOptions() instanceof OpenAiChatOptions defaults
+                && defaults.getModel() != null
+                && defaults.getModel().startsWith("doubao-seed-2-0-")) {
+            return new Prompt(promptText, OpenAiChatOptions.builder()
+                    .reasoningEffort("minimal")
+                    .maxTokens(CLASSIFICATION_MAX_TOKENS)
+                    .build());
+        }
+        return new Prompt(promptText);
+    }
+
     /**
      * 校验意图类型是否有效（模型可能返回枚举外的值）
      */

src/test/java/com/wok/supportbot/IntentRouterTests.java

@@ -0,0 +1,126 @@
+package com.wok.supportbot;
+
+import com.wok.supportbot.config.ChatModelFactory;
+import com.wok.supportbot.service.IntentRouter;
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Test;
+import org.mockito.ArgumentCaptor;
+import org.springframework.ai.chat.messages.AssistantMessage;
+import org.springframework.ai.chat.model.ChatModel;
+import org.springframework.ai.chat.model.ChatResponse;
+import org.springframework.ai.chat.model.Generation;
+import org.springframework.ai.openai.OpenAiChatModel;
+import org.springframework.ai.chat.prompt.Prompt;
+import org.springframework.ai.openai.api.OpenAiApi;
+import org.springframework.http.MediaType;
+import org.springframework.test.web.client.MockRestServiceServer;
+import org.springframework.web.client.RestClient;
+import org.springframework.ai.openai.OpenAiChatOptions;
+import org.springframework.test.util.ReflectionTestUtils;
+
+import java.util.List;
+
+import static org.junit.jupiter.api.Assertions.*;
+import static org.mockito.ArgumentMatchers.any;
+import static org.springframework.test.web.client.match.MockRestRequestMatchers.jsonPath;
+import static org.springframework.test.web.client.match.MockRestRequestMatchers.requestTo;
+import static org.springframework.test.web.client.response.MockRestResponseCreators.withSuccess;
+import static org.mockito.Mockito.*;
+
+class IntentRouterTests {
+    private final ChatModelFactory factory = mock(ChatModelFactory.class);
+    private final ChatModel model = mock(ChatModel.class);
+    private final IntentRouter router = new IntentRouter();
+
+    @BeforeEach
+    void setUp() {
+        ReflectionTestUtils.setField(router, "chatModelFactory", factory);
+        when(factory.getChatModel("CHAT")).thenReturn(model);
+    }
+
+    @Test
+    void seedClassificationDisablesThinkingWithoutChangingAnswerDefaults() {
+        OpenAiChatOptions defaults = OpenAiChatOptions.builder()
+                .model("doubao-seed-2-0-mini-260428")
+                .reasoningEffort("medium").maxTokens(2000).temperature(0.5).build();
+        when(model.getDefaultOptions()).thenReturn(defaults);
+        when(model.call(any(Prompt.class))).thenReturn(response("{\"intent\":\"RAG\",\"confidence\":0.95}"));
+
+        IntentRouter.IntentResult result = router.route("打印小票的排版规则是什么");
+
+        assertEquals("RAG", result.getIntent());
+        assertEquals(0.95, result.getConfidence());
+        ArgumentCaptor<Prompt> prompt = ArgumentCaptor.forClass(Prompt.class);
+        verify(model).call(prompt.capture());
+        OpenAiChatOptions options = assertInstanceOf(OpenAiChatOptions.class, prompt.getValue().getOptions());
+        assertEquals("minimal", options.getReasoningEffort());
+        assertEquals(128, options.getMaxTokens());
+        assertTrue(prompt.getValue().getContents().contains("打印小票的排版规则是什么"));
+        assertEquals("medium", defaults.getReasoningEffort());
+        assertEquals(2000, defaults.getMaxTokens());
+        assertEquals(0.5, defaults.getTemperature());
+    }
+
+    @Test
+    void otherModelsDoNotReceiveSeedSpecificOptions() {
+        when(model.getDefaultOptions()).thenReturn(OpenAiChatOptions.builder()
+                .model("deepseek-v4-flash").maxTokens(2000).build());
+        when(model.call(any(Prompt.class))).thenReturn(response("{\"intent\":\"CHITCHAT\",\"confidence\":0.9}"));
+
+        assertEquals("CHITCHAT", router.route("谢谢你的帮助").getIntent());
+
+        ArgumentCaptor<Prompt> prompt = ArgumentCaptor.forClass(Prompt.class);
+        verify(model).call(prompt.capture());
+        assertNull(prompt.getValue().getOptions());
+    }
+
+    @Test
+    void invalidOrTruncatedClassificationStillFallsBackToRetrieval() {
+        when(model.getDefaultOptions()).thenReturn(OpenAiChatOptions.builder()
+                .model("doubao-seed-2-0-mini-260428").build());
+        when(model.call(any(Prompt.class))).thenReturn(response("{\"intent\":"));
+
+        assertEquals("RAG", router.route("如何办理退款").getIntent());
+    }
+
+    @Test
+    void actualOpenAiRequestSendsFastClassificationOptions() {
+        RestClient.Builder restClient = RestClient.builder();
+        MockRestServiceServer server = MockRestServiceServer.bindTo(restClient).build();
+        OpenAiChatModel realModel = OpenAiChatModel.builder()
+                .openAiApi(OpenAiApi.builder().baseUrl("http://localhost").apiKey("test-key")
+                        .restClientBuilder(restClient).build())
+                .defaultOptions(OpenAiChatOptions.builder().model("doubao-seed-2-0-mini-260428")
+                        .maxTokens(2000).temperature(0.5).build())
+                .build();
+        when(factory.getChatModel("CHAT")).thenReturn(realModel);
+        server.expect(requestTo("http://localhost/v1/chat/completions"))
+                .andExpect(jsonPath("$.model").value("doubao-seed-2-0-mini-260428"))
+                .andExpect(jsonPath("$.reasoning_effort").value("minimal"))
+                .andExpect(jsonPath("$.max_tokens").value(128))
+                .andExpect(jsonPath("$.temperature").value(0.5))
+                .andRespond(withSuccess("""
+                        {"id":"classification","object":"chat.completion","created":1,
+                         "model":"doubao-seed-2-0-mini-260428",
+                         "choices":[{"index":0,"message":{"role":"assistant",
+                         "content":"{\\"intent\\":\\"RAG\\",\\"confidence\\":0.9}"},"finish_reason":"stop"}]}
+                        """, MediaType.APPLICATION_JSON));
+
+        IntentRouter.IntentResult result = router.route("如何办理退款");
+        assertEquals("RAG", result.getIntent());
+        assertEquals(0.9, result.getConfidence());
+        server.verify();
+        assertEquals(2000, realModel.getDefaultOptions().getMaxTokens());
+    }
+
+    @Test
+    void blankQuestionDoesNotCallModel() {
+        assertEquals("RAG", router.route(" ").getIntent());
+        verifyNoInteractions(model);
+        verify(factory, never()).getChatModel(any());
+    }
+
+    private ChatResponse response(String text) {
+        return new ChatResponse(List.of(new Generation(new AssistantMessage(text))));
+    }
+}