Preserve LLM context length for Flutter validation

Author: DevDesai444

sdk/runanywhere-flutter/packages/runanywhere/lib/native/dart_bridge_llm.dart

@@ -48,6 +48,7 @@ class DartBridgeLLM {
 
   RacHandle? _handle;
   String? _loadedModelId;
+  int? _loadedContextLength;
   final _logger = SDKLogger('DartBridge.LLM');
 
   /// Active stream subscription for cancellation
@@ -153,6 +154,7 @@ class DartBridgeLLM {
     String modelPath,
     String modelId,
     String modelName,
+    int? contextLength,
   ) async {
     final handle = getHandle();
 
@@ -181,6 +183,7 @@ class DartBridgeLLM {
       }
 
       _loadedModelId = modelId;
+      _loadedContextLength = contextLength;
       _logger.info('LLM model loaded: $modelId');
     } finally {
       calloc.free(pathPtr);
@@ -200,6 +203,7 @@ class DartBridgeLLM {
 
       cleanupFn(_handle!);
       _loadedModelId = null;
+      _loadedContextLength = null;
       _logger.info('LLM model unloaded');
     } catch (e) {
       _logger.error('Failed to unload LLM model: $e');
@@ -386,8 +390,10 @@ class DartBridgeLLM {
     String? systemPrompt,
     bool streamingEnabled = false,
   }) {
+    final contextLength = _loadedContextLength;
+
     LLMConfiguration(
-      contextLength: 32768,
+      contextLength: contextLength ?? 32768,
       maxTokens: maxTokens,
       temperature: temperature,
       systemPrompt: systemPrompt,
@@ -408,6 +414,7 @@ class DartBridgeLLM {
         destroyFn(_handle!);
         _handle = null;
         _loadedModelId = null;
+        _loadedContextLength = null;
         _logger.debug('LLM component destroyed');
       } catch (e) {
         _logger.error('Failed to destroy LLM component: $e');

sdk/runanywhere-flutter/packages/runanywhere/lib/native/dart_bridge_model_registry.dart

@@ -151,6 +151,7 @@ class DartBridgeModelRegistry {
         modelPtr.ref.localPath =
             pathDart != null ? strdupFn(pathDart) : nullptr;
         modelPtr.ref.downloadSize = model.sizeBytes;
+        modelPtr.ref.contextLength = model.contextLength;
         modelPtr.ref.source = model.source;
 
         final result = saveFn(_registryHandle!, modelPtr);
@@ -196,6 +197,7 @@ class DartBridgeModelRegistry {
         framework: _frameworkToFfi(model.framework),
         source: _sourceToFfi(model.source),
         sizeBytes: model.downloadSize ?? 0,
+        contextLength: model.contextLength ?? 0,
         downloadURL: model.downloadURL?.toString(),
         localPath: model.localPath?.toFilePath(),
         version: null,
@@ -384,6 +386,7 @@ class DartBridgeModelRegistry {
           ? Uri.file(ffiModel.localPath!)
           : null,
       downloadSize: ffiModel.sizeBytes > 0 ? ffiModel.sizeBytes : null,
+      contextLength: ffiModel.contextLength > 0 ? ffiModel.contextLength : null,
       source: _sourceFromFfi(ffiModel.source),
     );
   }
@@ -804,6 +807,7 @@ class DartBridgeModelRegistry {
       framework: struct.ref.framework,
       source: struct.ref.source,
       sizeBytes: struct.ref.downloadSize,
+      contextLength: struct.ref.contextLength,
       downloadURL: downloadURL,
       localPath: localPath,
       version: null,
@@ -1092,6 +1096,7 @@ class ModelInfo {
   final int framework;
   final int source;
   final int sizeBytes;
+  final int contextLength;
   final String? downloadURL;
   final String? localPath;
   final String? version;
@@ -1104,6 +1109,7 @@ class ModelInfo {
     required this.framework,
     required this.source,
     required this.sizeBytes,
+    required this.contextLength,
     this.downloadURL,
     this.localPath,
     this.version,
@@ -1119,6 +1125,7 @@ class ModelInfo {
         'framework': framework,
         'source': source,
         'sizeBytes': sizeBytes,
+        'contextLength': contextLength,
         if (downloadURL != null) 'downloadURL': downloadURL,
         if (localPath != null) 'localPath': localPath,
         if (version != null) 'version': version,

sdk/runanywhere-flutter/packages/runanywhere/lib/public/runanywhere.dart

@@ -451,7 +451,12 @@ class RunAnywhere {
       // Load model directly via DartBridgeLLM (mirrors Swift CppBridge.LLM pattern)
       // The C++ layer handles finding the right backend via the service registry
       logger.debug('Loading model via C++ bridge: $resolvedPath');
-      await DartBridge.llm.loadModel(resolvedPath, modelId, model.name);
+      await DartBridge.llm.loadModel(
+        resolvedPath,
+        modelId,
+        model.name,
+        model.contextLength,
+      );
 
       // Verify the model loaded successfully
       if (!DartBridge.llm.isLoaded) {
@@ -1943,7 +1948,7 @@ class RunAnywhere {
         latencyMs: latencyMs.round(),
         temperature: opts.temperature,
         maxTokens: opts.maxTokens,
-        contextLength: 8192, // Default context length for LlamaCpp
+        contextLength: modelInfo?.contextLength,
         tokensPerSecond: tokensPerSecond,
         isStreaming: false,
       );
@@ -2124,7 +2129,7 @@ class RunAnywhere {
         latencyMs: latencyMs.round(),
         temperature: opts.temperature,
         maxTokens: opts.maxTokens,
-        contextLength: 8192, // Default context length for LlamaCpp
+        contextLength: modelInfo?.contextLength,
         tokensPerSecond: tokensPerSecond,
         timeToFirstTokenMs: timeToFirstTokenMs,
         isStreaming: true,