feat: Introduce reasoning capabilities in chat and text completion

- Added AbstractCompletionRequest and AbstractCompletionResponse classes to standardize request and response structures.
- Implemented StreamOptions for streaming responses in chat completions.
- Enhanced ChatCompletionClient and TextCompletionClient to support reasoning content in responses.
- Updated ModelProfile to include reasoning effort and summary details.
- Created GptOssPromptStrategy to handle specific prompt formatting and reasoning extraction.
- Modified existing tests to validate reasoning content in completions.
- Updated configuration properties for reasoning settings in local properties files.

src/test/java/de/extio/lmlib/agent/AgentTest.java

@@ -29,7 +29,7 @@ import de.extio.lmlib.profile.ModelCategory;
 import io.netty.util.internal.ThreadLocalRandom;
 
 /**
- * These tests have been verified with Llama 3.1 8B and Gemma2 27B models
+ * These tests have been verified with Llama 3.1 8B, Gemma2 27B, Gemma3 27B and GPT-OSS models
  */
 @Disabled("This test requires a running Llama server or an Azure subscription (setup azure key in model profile)")
 @SpringBootTest(webEnvironment = WebEnvironment.NONE)
@@ -137,11 +137,14 @@ public class AgentTest {
 								null,
 								null,
 								null,
-								context -> {
-									if ("summary".equals(context.getStringValue(TextAgentResponseHandler.UPDATE_KEY))) { // UPDATE_KEY helps you to identify the streamed chunk
-										System.out.print(context.getStringValue("summary_chunk")); // Single chunks can be accessed via _chunk suffix, depends on the StreamedAgentResponseHandler implementation
-										lastStreamedSummary.set(context.getStringValue("summary")); // The current key value can also be accessed (e.g. concatenated, depends on the StreamedAgentResponseHandler implementation)
-									}									
+								(context, chunk) -> {
+									if (chunk.content() != null) { // Since chunk can contain both content and reasoning content, we check for null
+										context.setStringValue("summary_chunk", chunk.content());
+									}
+									if (chunk.reasoningContent() != null) {
+										// System.out.print(chunk.reasoningContent()); // The reasoning can also be accessed if available
+									}
+									lastStreamedSummary.set(context.getStringValue("summary")); // The current key value can also be accessed (e.g. concatenated, depends on the StreamedAgentResponseHandler implementation)
 								}
 						),
 						null,
@@ -157,6 +160,7 @@ public class AgentTest {
 		
 		final var resultContexts = this.agentExecutor.walk(agents.get("CodeSummarizer"), context, contextUpdate -> {
 			// This is an optional consumer that can be used to receive updates from the agent context
+			System.out.print(contextUpdate.getStringValue("summary_chunk"));
 		});
 		
 		assertEquals(1, resultContexts.size());

src/test/java/de/extio/lmlib/client/oai/CompletionClientTest.java

@@ -1,6 +1,7 @@
 package de.extio.lmlib.client.oai;
 
 import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertTrue;
 
 import java.time.LocalDateTime;
@@ -45,10 +46,11 @@ public class CompletionClientTest {
 	@Test
 	void completion() throws Exception {
 		final var completion = this.textCompletionClient.completion(
-				null,
+				ModelCategory.SMALL,
 				"You are a helpful assistant",
 				"Say the color is green");
 		
+		LOGGER.info(completion.reasoning());
 		LOGGER.info(completion.response());
 		
 		assertTrue(Grader.assessScoreBinary("Does the text mention the color green?", completion.response(), this.chatCompletionclient));
@@ -61,6 +63,7 @@ public class CompletionClientTest {
 		conversation.addTurn(new Turn(TurnType.USER, "You choose!"));
 		final var completion = this.textCompletionClient.conversation(ModelCategory.SMALL, conversation);
 		
+		LOGGER.info(completion.reasoning());
 		LOGGER.info(completion.response());
 		
 		assertTrue(Grader.assessScoreBinary("Does the text mention a color?", completion.response(), this.chatCompletionclient));
@@ -73,11 +76,24 @@ public class CompletionClientTest {
 		conversation.addTurn(new Turn(TurnType.USER, "You choose! But I need you to explain the character attributes to me."));
 		final StringBuilder sb = new StringBuilder();
 		final var completion = this.textCompletionClient.streamConversation(ModelCategory.SMALL, conversation, chunk -> {
-			System.out.print(chunk);
-			sb.append(chunk);
+			if (chunk.reasoningContent() != null) {
+				System.out.print(chunk.reasoningContent());
+			}
+			if (chunk.content() != null) {
+				System.out.print(chunk.content());
+				sb.append(chunk.content());
+			}
 		});
 		
-		assertEquals(completion.response(), sb.toString());
+		if (completion.reasoning() != null) {
+			// Prompt template does not separate reasoning and response from the stream at the moment
+			assertTrue(sb.toString().contains(completion.reasoning()));
+			assertTrue(sb.toString().contains(completion.response()));
+			assertFalse(completion.response().contains(completion.reasoning()));
+		}
+		else {
+			assertEquals(completion.response(), sb.toString());
+		}
 		assertTrue(Grader.assessScoreBinary("Does the text mention a role-playing game character?", completion.response(), this.chatCompletionclient));
 	}
 	
@@ -88,6 +104,7 @@ public class CompletionClientTest {
 		conversation.addTurn(new Turn(TurnType.USER, "You choose!"));
 		final var completion = this.chatCompletionclient.conversation(ModelCategory.MEDIUM, conversation);
 		
+		LOGGER.info(completion.reasoning());
 		LOGGER.info(completion.response());
 		
 		assertTrue(Grader.assessScoreBinary("Does the text mention a color?", completion.response(), this.chatCompletionclient));
@@ -100,8 +117,13 @@ public class CompletionClientTest {
 		conversation.addTurn(new Turn(TurnType.USER, "You choose! But I need you to explain the opening to me."));
 		final StringBuilder sb = new StringBuilder();
 		final var completion = this.chatCompletionclient.streamConversation(ModelCategory.MEDIUM, conversation, chunk -> {
-			System.out.print(chunk);
-			sb.append(chunk);
+			if (chunk.reasoningContent() != null) {
+				System.out.print(chunk.reasoningContent());
+			}
+			if (chunk.content() != null) {
+				System.out.print(chunk.content());
+				sb.append(chunk.content());
+			}
 		});
 		
 		assertEquals(completion.response(), sb.toString());

src/test/resources/local-chatcompletion.properties

@@ -1,7 +1,7 @@
 prompts=
 tokenEncoding=cl100k_base
-maxTokens=2000
-maxContextLength=8000
+maxTokens=2500
+maxContextLength=16000
 temperature=0.4
 topP=1.0
 modelProvider=OAI_CHAT_COMPLETION
@@ -10,3 +10,5 @@ url=http://localhost:5001
 apiKey=
 cost1MInTokens=0
 cost1MOutTokens=0
+reasoningEffort=medium
+reasoningSummaryDetails=auto

src/test/resources/local-completion.properties

@@ -1,7 +1,7 @@
-prompts=gemma
+prompts=gpt-oss
 tokenEncoding=cl100k_base
-maxTokens=2000
-maxContextLength=8000
+maxTokens=2500
+maxContextLength=16000
 temperature=0.4
 topP=1.0
 modelProvider=OAI_TEXT_COMPLETION