Merge pull request #117 from IBM/add-enabled-field-to-thinking

Add explicit `enabled` flag to Thinking configuration

Author: andreadimaio

modules/watsonx-ai/src/main/java/com/ibm/watsonx/ai/chat/ChatRequest.java

@@ -4,6 +4,7 @@
  */
 package com.ibm.watsonx.ai.chat;
 
+import static java.util.Objects.isNull;
 import static java.util.Objects.requireNonNull;
 import java.util.List;
 import com.ibm.watsonx.ai.chat.model.ChatMessage;
@@ -198,12 +199,7 @@ public Builder parameters(ChatParameters parameters) {
          * @param enabled {@code true} to enable reasoning with default settings, {@code false} to disable reasoning
          */
         public Builder thinking(boolean enabled) {
-            if (enabled) {
-                return thinking(Thinking.builder().build());
-            } else {
-                thinking = null;
-                return this;
-            }
+            return thinking(Thinking.builder().enabled(enabled).build());
         }
 
         /**
@@ -223,6 +219,11 @@ public Builder thinking(boolean enabled) {
          * @param tags an {@link ExtractionTags} instance defining the reasoning and response tags
          */
         public Builder thinking(ExtractionTags tags) {
+            if (isNull(tags)) {
+                thinking = null;
+                return this;
+            }
+
             return thinking(Thinking.of(tags));
         }
 
@@ -242,6 +243,11 @@ public Builder thinking(ExtractionTags tags) {
          * @param thinkingEffort the desired {@link ThinkingEffort} level
          */
         public Builder thinking(ThinkingEffort thinkingEffort) {
+            if (isNull(thinkingEffort)) {
+                thinking = null;
+                return this;
+            }
+
             return thinking(Thinking.of(thinkingEffort));
         }
 

modules/watsonx-ai/src/main/java/com/ibm/watsonx/ai/chat/ChatService.java

@@ -93,10 +93,11 @@ public ChatResponse chat(ChatRequest chatRequest) {
         Map<String, Object> chatTemplateKwargs = null;
         if (nonNull(chatRequest.getThinking())) {
             var thinking = chatRequest.getThinking();
-            chatTemplateKwargs = Map.of("thinking", true);
             extractionTags = thinking.getExtractionTags();
             includeReasoning = thinking.getIncludeReasoning();
             thinkingEffort = nonNull(thinking.getThinkingEffort()) ? thinking.getThinkingEffort().getValue() : null;
+            if (nonNull(thinking.getEnabled()))
+                chatTemplateKwargs = Map.of("thinking", thinking.getEnabled());
         }
 
         var textChatRequest = TextChatRequest.builder()
@@ -153,10 +154,11 @@ public CompletableFuture<Void> chatStreaming(ChatRequest chatRequest, ChatHandle
         Map<String, Object> chatTemplateKwargs = null;
         if (nonNull(chatRequest.getThinking())) {
             var thinking = chatRequest.getThinking();
-            chatTemplateKwargs = Map.of("thinking", true);
             extractionTags = thinking.getExtractionTags();
             includeReasoning = thinking.getIncludeReasoning();
             thinkingEffort = nonNull(thinking.getThinkingEffort()) ? thinking.getThinkingEffort().getValue() : null;
+            if (nonNull(thinking.getEnabled()))
+                chatTemplateKwargs = Map.of("thinking", thinking.getEnabled());
         }
 
         var textChatRequest = TextChatRequest.builder()

modules/watsonx-ai/src/main/java/com/ibm/watsonx/ai/chat/model/Thinking.java

@@ -4,6 +4,9 @@
  */
 package com.ibm.watsonx.ai.chat.model;
 
+import static java.util.Objects.isNull;
+import static java.util.Objects.nonNull;
+
 /**
  * Represents the reasoning configuration used by the LLM.
  * <p>
@@ -35,16 +38,25 @@
  */
 public final class Thinking {
 
+    private final Boolean enabled;
     private final Boolean includeReasoning;
     private final ExtractionTags extractionTags;
     private final ThinkingEffort thinkingEffort;
 
     private Thinking(Builder builder) {
+        enabled = builder.enabled;
         includeReasoning = builder.includeReasoning;
         extractionTags = builder.extractionTags;
         thinkingEffort = builder.thinkingEffort;
     }
 
+    public Boolean getEnabled() {
+        if (isNull(enabled) && (nonNull(includeReasoning) || nonNull(extractionTags) || nonNull(thinkingEffort)))
+            return true;
+
+        return enabled;
+    }
+
     public Boolean getIncludeReasoning() {
         return includeReasoning;
     }
@@ -75,12 +87,19 @@ public static Builder builder() {
     }
 
     public final static class Builder {
+        private Boolean enabled;
         private Boolean includeReasoning;
         private ExtractionTags extractionTags;
         private ThinkingEffort thinkingEffort;
 
         private Builder() {}
 
+
+        public Builder enabled(Boolean enabled) {
+            this.enabled = enabled;
+            return this;
+        }
+
         /**
          * Sets whether reasoning should be included in the LLM response.
          * <p>

modules/watsonx-ai/src/test/java/com/ibm/watsonx/ai/chat/ChatRequestTest.java

@@ -0,0 +1,36 @@
+/*
+ * Copyright IBM Corp. 2025 - 2025
+ * SPDX-License-Identifier: Apache-2.0
+ */
+package com.ibm.watsonx.ai.chat;
+
+import static org.junit.jupiter.api.Assertions.assertNull;
+import org.junit.jupiter.api.Test;
+import com.ibm.watsonx.ai.chat.model.ExtractionTags;
+import com.ibm.watsonx.ai.chat.model.ThinkingEffort;
+import com.ibm.watsonx.ai.chat.model.UserMessage;
+
+public class ChatRequestTest {
+
+    @Test
+    void should_not_set_thinking_when_null_is_passed() {
+
+        ExtractionTags tags = null;
+
+        var chatRequest = ChatRequest.builder()
+            .messages(UserMessage.text("Hello"))
+            .thinking(tags)
+            .build();
+
+        assertNull(chatRequest.getThinking());
+
+        ThinkingEffort thinkingEffort = null;
+
+        chatRequest = ChatRequest.builder()
+            .messages(UserMessage.text("Hello"))
+            .thinking(thinkingEffort)
+            .build();
+
+        assertNull(chatRequest.getThinking());
+    }
+}

modules/watsonx-ai/src/test/java/com/ibm/watsonx/ai/chat/ChatServiceThinkingTest.java

@@ -249,7 +249,7 @@ void test_thinking_with_efforts() throws Exception {
         }
 
         @Test
-        void test_thinking_without_thinking() throws Exception {
+        void test_thinking_with_enabled_false() throws Exception {
 
             wireMock.stubFor(post("/ml/v1/text/chat?version=%s".formatted(API_VERSION))
                 .withHeader("Authorization", equalTo("Bearer my-token"))
@@ -268,7 +268,10 @@ void test_thinking_without_thinking() throws Exception {
                                 ]
                             }
                         ],
-                        "time_limit": 60000
+                        "time_limit": 60000,
+                        "chat_template_kwargs" : {
+                            "thinking": false
+                        }
                     }"""))
                 .willReturn(aResponse()
                     .withStatus(200)
@@ -798,7 +801,7 @@ public void onPartialThinking(String partialThinking, PartialChatResponse partia
         }
 
         @Test
-        void test_thinking_without_thinking() throws Exception {
+        void test_thinking_with_enabled_false() throws Exception {
 
             wireMock.stubFor(post("/ml/v1/text/chat_stream?version=%s".formatted(API_VERSION))
                 .withHeader("Authorization", equalTo("Bearer my-token"))
@@ -898,7 +901,6 @@ void test_thinking_without_thinking() throws Exception {
                 .build();
 
             var chatRequest = ChatRequest.builder()
-                .thinking(false)
                 .messages(UserMessage.text("Translate 'Hello' to Italian"))
                 .build();