Introduce thought configuration to quarkus-langchain4j-ai-gemini

Closes: #1694

Author: geoand

model-providers/google/gemini/ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/ai/runtime/gemini/AiGeminiChatLanguageModel.java

@@ -25,7 +25,7 @@ public class AiGeminiChatLanguageModel extends GeminiChatLanguageModel {
 
     private AiGeminiChatLanguageModel(Builder builder) {
         super(builder.modelId, builder.temperature, builder.maxOutputTokens, builder.topK, builder.topP, builder.responseFormat,
-                builder.listeners);
+                builder.listeners, builder.thinkingBudget, builder.includeThoughts);
 
         this.apiMetadata = AiGeminiRestApi.ApiMetadata
                 .builder()
@@ -81,6 +81,8 @@ public static final class Builder {
         private Boolean logRequests = false;
         private Boolean logResponses = false;
         private List<ChatModelListener> listeners = Collections.emptyList();
+        private Long thinkingBudget;
+        private boolean includeThoughts = false;
 
         public Builder configName(String configName) {
             this.configName = configName;
@@ -147,6 +149,16 @@ public Builder listeners(List<ChatModelListener> listeners) {
             return this;
         }
 
+        public Builder thinkingBudget(Long thinkingBudget) {
+            this.thinkingBudget = thinkingBudget;
+            return this;
+        }
+
+        public Builder includeThoughts(Boolean includeThoughts) {
+            this.includeThoughts = includeThoughts;
+            return this;
+        }
+
         public AiGeminiChatLanguageModel build() {
             return new AiGeminiChatLanguageModel(this);
         }

model-providers/google/gemini/ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/ai/runtime/gemini/AiGeminiRecorder.java

@@ -16,6 +16,7 @@
 import dev.langchain4j.model.chat.listener.ChatModelListener;
 import dev.langchain4j.model.embedding.DisabledEmbeddingModel;
 import dev.langchain4j.model.embedding.EmbeddingModel;
+import io.quarkiverse.langchain4j.ai.runtime.gemini.config.ChatModelConfig;
 import io.quarkiverse.langchain4j.ai.runtime.gemini.config.LangChain4jAiGeminiConfig;
 import io.quarkiverse.langchain4j.auth.ModelAuthProvider;
 import io.quarkiverse.langchain4j.runtime.NamedConfigUtil;
@@ -109,6 +110,13 @@ public Function<SyntheticCreationalContext<ChatModel>, ChatModel> chatModel(Stri
             if (chatModelConfig.timeout().isPresent()) {
                 builder.timeout(chatModelConfig.timeout().get());
             }
+            ChatModelConfig.ThinkingConfig thinkingConfig = chatModelConfig.thinking();
+            if (thinkingConfig.includeThoughts()) {
+                builder.includeThoughts(thinkingConfig.includeThoughts());
+            }
+            if (thinkingConfig.thinkingBudget().isPresent()) {
+                builder.thinkingBudget(thinkingConfig.thinkingBudget().getAsLong());
+            }
 
             // TODO: add the rest of the properties
 

model-providers/google/gemini/ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/ai/runtime/gemini/config/ChatModelConfig.java

@@ -4,7 +4,9 @@
 import java.util.Optional;
 import java.util.OptionalDouble;
 import java.util.OptionalInt;
+import java.util.OptionalLong;
 
+import io.quarkiverse.langchain4j.gemini.common.ThinkingConfig;
 import io.quarkus.runtime.annotations.ConfigDocDefault;
 import io.quarkus.runtime.annotations.ConfigGroup;
 import io.smallrye.config.WithDefault;
@@ -93,4 +95,33 @@ public interface ChatModelConfig {
     @WithDefault("${quarkus.langchain4j.ai.gemini.timeout}")
     Optional<Duration> timeout();
 
+    /**
+     * Thought related configuration
+     */
+    ThinkingConfig thinking();
+
+    interface ThinkingConfig {
+
+        /**
+         * Controls whether thought summaries are enabled.
+         * Thought summaries are synthesized versions of the model's raw thoughts and offer insights into the model's internal
+         * reasoning process.
+         */
+        @WithDefault("false")
+        boolean includeThoughts();
+
+        /**
+         * The thinkingBudget parameter guides the model on the number of thinking tokens to use when generating a response.
+         * A higher token count generally allows for more detailed reasoning, which can be beneficial for tackling more complex
+         * tasks.
+         * If latency is more important, use a lower budget or disable thinking by setting thinkingBudget to 0.
+         * Setting the thinkingBudget to -1 turns on dynamic thinking, meaning the model will adjust the budget based on the
+         * complexity of the request.
+         * <p>
+         * The thinkingBudget is only supported in Gemini 2.5 Flash, 2.5 Pro, and 2.5 Flash-Lite. Depending on the prompt, the
+         * model might overflow or underflow the token budget.
+         * See <a href="https://ai.google.dev/gemini-api/docs/thinking#set-budget">Gemini API docs</a> for more details.
+         */
+        OptionalLong thinkingBudget();
+    }
 }

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/BaseGeminiChatModel.java

@@ -17,15 +17,20 @@ public class BaseGeminiChatModel {
     protected final Double topP;
     protected final ResponseFormat responseFormat;
     protected final List<ChatModelListener> listeners;
+    protected final Long thinkingBudget;
+    protected final boolean includeThoughts;
 
     public BaseGeminiChatModel(String modelId, Double temperature, Integer maxOutputTokens, Integer topK, Double topP,
-            ResponseFormat responseFormat, List<ChatModelListener> listeners) {
+            ResponseFormat responseFormat, List<ChatModelListener> listeners, Long thinkingBudget,
+            boolean includeThoughts) {
         this.modelId = modelId;
         this.temperature = temperature;
         this.maxOutputTokens = maxOutputTokens;
         this.topK = topK;
         this.topP = topP;
         this.responseFormat = responseFormat;
         this.listeners = listeners;
+        this.thinkingBudget = thinkingBudget;
+        this.includeThoughts = includeThoughts;
     }
 }

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/GeminiChatLanguageModel.java

@@ -36,8 +36,9 @@
 public abstract class GeminiChatLanguageModel extends BaseGeminiChatModel implements ChatModel {
 
     public GeminiChatLanguageModel(String modelId, Double temperature, Integer maxOutputTokens, Integer topK,
-            Double topP, ResponseFormat responseFormat, List<ChatModelListener> listeners) {
-        super(modelId, temperature, maxOutputTokens, topK, topP, responseFormat, listeners);
+            Double topP, ResponseFormat responseFormat, List<ChatModelListener> listeners, Long thinkingBudget,
+            boolean includeThoughts) {
+        super(modelId, temperature, maxOutputTokens, topK, topP, responseFormat, listeners, thinkingBudget, includeThoughts);
     }
 
     @Override
@@ -56,7 +57,7 @@ public Set<Capability> supportedCapabilities() {
     public ChatResponse chat(ChatRequest chatRequest) {
         ChatRequestParameters requestParameters = chatRequest.parameters();
         ResponseFormat effectiveResponseFormat = getOrDefault(requestParameters.responseFormat(), responseFormat);
-        GenerationConfig generationConfig = GenerationConfig.builder()
+        GenerationConfig.Builder generationConfigBuilder = GenerationConfig.builder()
                 .maxOutputTokens(getOrDefault(requestParameters.maxOutputTokens(), this.maxOutputTokens))
                 .responseMimeType(computeMimeType(effectiveResponseFormat))
                 .responseSchema(effectiveResponseFormat != null
@@ -65,8 +66,11 @@ public ChatResponse chat(ChatRequest chatRequest) {
                 .stopSequences(requestParameters.stopSequences())
                 .temperature(getOrDefault(requestParameters.temperature(), this.temperature))
                 .topK(getOrDefault(requestParameters.topK(), this.topK))
-                .topP(getOrDefault(requestParameters.topP(), this.topP))
-                .build();
+                .topP(getOrDefault(requestParameters.topP(), this.topP));
+        if (includeThoughts) {
+            generationConfigBuilder.thinkingConfig(new ThinkingConfig(thinkingBudget, includeThoughts));
+        }
+        GenerationConfig generationConfig = generationConfigBuilder.build();
         GenerateContentRequest request = ContentMapper.map(chatRequest.messages(), chatRequest.toolSpecifications(),
                 generationConfig);
 

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/GeminiStreamingChatLanguageModel.java

@@ -36,7 +36,7 @@ public abstract class GeminiStreamingChatLanguageModel extends BaseGeminiChatMod
 
     public GeminiStreamingChatLanguageModel(String modelId, Double temperature, Integer maxOutputTokens, Integer topK,
             Double topP, ResponseFormat responseFormat, List<ChatModelListener> listeners) {
-        super(modelId, temperature, maxOutputTokens, topK, topP, responseFormat, listeners);
+        super(modelId, temperature, maxOutputTokens, topK, topP, responseFormat, listeners, null, false);
     }
 
     @Override

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/GenerateContentResponse.java

@@ -11,7 +11,7 @@ public record Content(List<Part> parts) {
 
         }
 
-        public record Part(String text, FunctionCall functionCall) {
+        public record Part(String text, FunctionCall functionCall, Boolean thought) {
 
         }
 

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/GenerationConfig.java

@@ -11,6 +11,7 @@ public class GenerationConfig {
     private final String responseMimeType;
     private final Schema responseSchema;
     private final List<String> stopSequences;
+    private final ThinkingConfig thinkingConfig;
 
     public GenerationConfig(Builder builder) {
         this.temperature = builder.temperature;
@@ -20,6 +21,7 @@ public GenerationConfig(Builder builder) {
         this.responseMimeType = builder.responseMimeType;
         this.responseSchema = builder.responseSchema;
         this.stopSequences = builder.stopSequences;
+        this.thinkingConfig = builder.thinkingConfig;
     }
 
     public Double getTemperature() {
@@ -50,6 +52,10 @@ public List<String> getStopSequences() {
         return stopSequences;
     }
 
+    public ThinkingConfig getThinkingConfig() {
+        return thinkingConfig;
+    }
+
     public static Builder builder() {
         return new Builder();
     }
@@ -63,6 +69,7 @@ public static final class Builder {
         private String responseMimeType;
         private Schema responseSchema;
         private List<String> stopSequences;
+        private ThinkingConfig thinkingConfig;
 
         public Builder temperature(Double temperature) {
             this.temperature = temperature;
@@ -99,6 +106,11 @@ public Builder stopSequences(List<String> stopSequences) {
             return this;
         }
 
+        public Builder thinkingConfig(ThinkingConfig thinkingConfig) {
+            this.thinkingConfig = thinkingConfig;
+            return this;
+        }
+
         public GenerationConfig build() {
             return new GenerationConfig(this);
         }

model-providers/google/gemini/gemini-common/runtime/src/main/java/io/quarkiverse/langchain4j/gemini/common/ThinkingConfig.java

@@ -0,0 +1,20 @@
+package io.quarkiverse.langchain4j.gemini.common;
+
+public class ThinkingConfig {
+
+    private final Long thinkingBudget;
+    private final Boolean includeThoughts;
+
+    public ThinkingConfig(Long thinkingBudget, Boolean includeThoughts) {
+        this.thinkingBudget = thinkingBudget;
+        this.includeThoughts = includeThoughts;
+    }
+
+    public Long getThinkingBudget() {
+        return thinkingBudget;
+    }
+
+    public Boolean getIncludeThoughts() {
+        return includeThoughts;
+    }
+}

model-providers/google/vertex-ai-gemini/runtime/src/main/java/io/quarkiverse/langchain4j/vertexai/runtime/gemini/VertexAiGeminiChatLanguageModel.java

@@ -25,7 +25,7 @@ public class VertexAiGeminiChatLanguageModel extends GeminiChatLanguageModel {
 
     private VertexAiGeminiChatLanguageModel(Builder builder) {
         super(builder.modelId, builder.temperature, builder.maxOutputTokens, builder.topK, builder.topP, builder.responseFormat,
-                builder.listeners);
+                builder.listeners, null, false);
 
         this.apiMetadata = VertxAiGeminiRestApi.ApiMetadata
                 .builder()