Add PRESERVE_THINKING and REASONING_EFFORT_PRESERVE_THINKING placeholders and update UI visibility

README.md

@@ -46,6 +46,8 @@ Use these tokens in app arguments:
 - `TOP_K`
 - `MIN_P`
 - `REASONING_EFFORT`
+- `PRESERVE_THINKING`
+- `REASONING_EFFORT_PRESERVE_THINKING`
 - `REASONING_STOP`
 
 Example:
@@ -56,8 +58,10 @@ Example:
 
 For llama-server reasoning controls, configure `--chat-template-kwargs REASONING_EFFORT`.
 The placeholder expands to a single JSON argument such as `{"reasoning_effort": "medium"}`.
-When this placeholder is selected for the current app and model, the model options panel provides
-the presets `none`, `low`, `medium`, `high`, and `xhigh`.
+`PRESERVE_THINKING` expands to `{"preserve_thinking": true}`, while
+`REASONING_EFFORT_PRESERVE_THINKING` combines both settings in one kwargs JSON argument.
+The model options panel always provides the reasoning-effort presets `none`, `low`, `medium`,
+`high`, and `xhigh`.
 
 ## Setup
 

src/de/extio/lm_launcher/App.java

@@ -18,17 +18,19 @@ record App(Path path, Path interpreter, @Deprecated List<String> arguments, List
 		modelSettings = modelSettings == null ? new ArrayList<>() : modelSettings;
 	}
 	
-	static Map<String, Function<Model, String>> PLACEHOLDERS = Map.of(
-			"MODEL", m -> m.path(),
-			"CONTEXT_SIZE", m -> String.valueOf(m.contextSize()),
-			"THREADS", m -> String.valueOf(m.threads()),
-			"GPU_LAYERS", m -> String.valueOf(m.gpuLayers()),
-			"TEMP", Model::temperatureDisplay,
-			"TOP_P", Model::topPDisplay,
-			"TOP_K", Model::topKDisplay,
-			"MIN_P", Model::minPDisplay,
-			"REASONING_EFFORT", Model::reasoningEffortArgument,
-			"REASONING_STOP", m -> "... thinking budget exceeded, let's answer now.\n");
+	static Map<String, Function<Model, String>> PLACEHOLDERS = Map.ofEntries(
+			Map.entry("MODEL", m -> m.path()),
+			Map.entry("CONTEXT_SIZE", m -> String.valueOf(m.contextSize())),
+			Map.entry("THREADS", m -> String.valueOf(m.threads())),
+			Map.entry("GPU_LAYERS", m -> String.valueOf(m.gpuLayers())),
+			Map.entry("TEMP", Model::temperatureDisplay),
+			Map.entry("TOP_P", Model::topPDisplay),
+			Map.entry("TOP_K", Model::topKDisplay),
+			Map.entry("MIN_P", Model::minPDisplay),
+			Map.entry("REASONING_EFFORT", Model::reasoningEffortArgument),
+			Map.entry("PRESERVE_THINKING", Model::preserveThinkingArgument),
+			Map.entry("REASONING_EFFORT_PRESERVE_THINKING", Model::reasoningEffortPreserveThinkingArgument),
+			Map.entry("REASONING_STOP", m -> "... thinking budget exceeded, let's answer now.\n"));
 	
 	String argumentsToString(final boolean showAll) {
 		return this.argumentsToStringByArgs(this.appArguments.stream(), showAll);

src/de/extio/lm_launcher/LaunchFrame.java

@@ -695,16 +695,8 @@ public class LaunchFrame extends JFrame {
 	}
 
 	private void updateReasoningEffortVisibility() {
-		boolean selected = false;
-		for (int i = 0; i < this.appArgumentsTableModel.getRowCount(); i++) {
-			final AppArgument argument = (AppArgument) this.appArgumentsTableModel.getValueAt(i, 0);
-			if ("REASONING_EFFORT".equals(argument.argument()) && (!argument.optional() || argument.default_())) {
-				selected = true;
-				break;
-			}
-		}
-		this.lblReasoningEffort.setVisible(selected);
-		this.comboReasoningEffort.setVisible(selected);
+		this.lblReasoningEffort.setVisible(true);
+		this.comboReasoningEffort.setVisible(true);
 		this.panel_3.revalidate();
 		this.panel_3.repaint();
 	}

src/de/extio/lm_launcher/Model.java

@@ -50,6 +50,14 @@ record Model(String path, int contextSize, int maxContextSize, int gpuLayers, in
 		return "{\"reasoning_effort\": \"" + this.reasoningEffort + "\"}";
 	}
 
+	String preserveThinkingArgument() {
+		return "{\"preserve_thinking\": true}";
+	}
+
+	String reasoningEffortPreserveThinkingArgument() {
+		return "{\"reasoning_effort\": \"" + this.reasoningEffort + "\", \"preserve_thinking\": true}";
+	}
+
 	private static boolean isValidReasoningEffort(final String value) {
 		for (final String effort : REASONING_EFFORTS) {
 			if (effort.equals(value)) {