Formatting

Author: orionpapadakis

src/main/java/org/beehive/gpullama3/model/ModelType.java

@@ -16,7 +16,7 @@
  * <p><b>Usage:</b> Use {@code ModelType} to specify or retrieve the type of
  * large language model (LLM), such as Llama or Qwen3. This ensures clean and structured handling of model behaviors and configurations by
  * dispatching calls to the appropriate model loader for each
- *  model type.</p>
+ * model type.</p>
  *
  * <p>Each enum value represents a distinct model type, which might be used for
  * conditional logic, initialization, or resource allocation within GPULlama3.java.</p>

src/main/java/org/beehive/gpullama3/model/loader/AbstractModelLoader.java

@@ -16,10 +16,8 @@
 /**
  * Abstract base class for model loaders using Template Method pattern. Provides common loading flow with extension points for model-specific logic.
  *
- * @param <M>
- *         The specific Model type to load
- * @param <C>
- *         The specific Configuration type for the model
+ * @param <M> The specific Model type to load
+ * @param <C> The specific Configuration type for the model
  */
 public abstract class AbstractModelLoader<M extends Model, C extends Configuration> {
 
@@ -77,39 +75,33 @@ public final M loadModel() {
     /**
      * Load the vocabulary from GGUF metadata. Model-specific implementations should override this method.
      *
-     * @param metadata
-     *         The GGUF metadata map
+     * @param metadata The GGUF metadata map
      * @return The loaded Vocabulary
      */
     protected abstract Vocabulary loadVocabulary(Map<String, Object> metadata);
 
     /**
      * Create a tokenizer instance for this model.
      *
-     * @param metadata
-     *         The GGUF metadata map
-     * @param vocabulary
-     *         The loaded vocabulary
+     * @param metadata   The GGUF metadata map
+     * @param vocabulary The loaded vocabulary
      * @return The tokenizer instance
      */
     protected abstract Tokenizer createTokenizer(Map<String, Object> metadata, Vocabulary vocabulary);
 
     /**
      * Create a configuration instance from GGUF metadata.
      *
-     * @param metadata
-     *         The GGUF metadata map
+     * @param metadata The GGUF metadata map
      * @return The configuration instance
      */
     protected abstract C createConfiguration(Map<String, Object> metadata);
 
     /**
      * Load model weights from tensor entries. Default implementation handles common weight loading logic.
      *
-     * @param tensorEntries
-     *         Map of tensor names to tensor entries
-     * @param config
-     *         The model configuration
+     * @param tensorEntries Map of tensor names to tensor entries
+     * @param config        The model configuration
      * @return The loaded weights
      */
     public Weights loadWeights(Map<String, GGMLTensorEntry> tensorEntries, C config) {
@@ -131,12 +123,9 @@ public Weights loadWeights(Map<String, GGMLTensorEntry> tensorEntries, C config)
     /**
      * Create the final model instance.
      *
-     * @param config
-     *         The model configuration
-     * @param tokenizer
-     *         The tokenizer
-     * @param weights
-     *         The loaded weights
+     * @param config    The model configuration
+     * @param tokenizer The tokenizer
+     * @param weights   The loaded weights
      * @return The model instance
      */
     protected abstract M createModel(C config, Tokenizer tokenizer, Weights weights);
@@ -164,11 +153,11 @@ protected GGMLTensorEntry getOutputWeight(Map<String, GGMLTensorEntry> tensorEnt
      * Create standard (CPU) weights.
      */
     protected abstract Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntries, C config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight);
+                                                     GGMLTensorEntry outputWeight);
 
     /**
      * Create TornadoVM (GPU) weights.
      */
     protected abstract Weights createTornadoVMWeights(Map<String, GGMLTensorEntry> tensorEntries, C config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight);
+                                                      GGMLTensorEntry outputWeight);
 }

src/main/java/org/beehive/gpullama3/model/loader/LlamaModelLoader.java

@@ -73,7 +73,7 @@ protected Llama createModel(LlamaConfiguration config, Tokenizer tokenizer, Weig
 
     @Override
     protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntries, LlamaConfiguration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                            GGMLTensorEntry outputWeight) {
 
         final int nl = config.numberOfLayers();
 

src/main/java/org/beehive/gpullama3/model/loader/MistralModelLoader.java

@@ -70,7 +70,7 @@ protected Mistral createModel(MistralConfiguration config, Tokenizer tokenizer,
 
     @Override
     protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntries, MistralConfiguration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                            GGMLTensorEntry outputWeight) {
 
         final int nl = config.numberOfLayers();
 

src/main/java/org/beehive/gpullama3/model/loader/ModelLoader.java

@@ -21,7 +21,6 @@
 import java.nio.FloatBuffer;
 import java.nio.channels.FileChannel;
 import java.nio.file.Path;
-import java.nio.file.StandardOpenOption;
 import java.util.Map;
 import java.util.Set;
 import java.util.function.IntFunction;
@@ -74,13 +73,10 @@ private static ModelType detectModelType(Map<String, Object> metadata) {
      * If Ahead-of-Time (AOT) mode is enabled, attempts to use a pre-loaded compiled model. Otherwise, loads the model from the specified path using the model loader.
      * </p>
      *
-     * @param options
-     *         the parsed CLI options containing model path and max token limit
+     * @param options the parsed CLI options containing model path and max token limit
      * @return the loaded {@link Model} instance
-     * @throws IOException
-     *         if the model fails to load
-     * @throws IllegalStateException
-     *         if AOT loading is enabled but the preloaded model is unavailable
+     * @throws IOException           if the model fails to load
+     * @throws IllegalStateException if AOT loading is enabled but the preloaded model is unavailable
      */
     public static Model loadModel(Options options) throws IOException {
         Path ggufPath = options.modelPath();
@@ -163,7 +159,7 @@ public static TornadoTensor loadTornadoTensorAsFP32(GGMLTensorEntry entry) {
                 HalfFloatArray tensorHFA = tensor.asHalfFloatArray();
                 int numOfElements = tensorHFA.getSize();
                 FloatArray tensorFA = new FloatArray(numOfElements);
-                for(int i = 0; i < numOfElements; i++) {
+                for (int i = 0; i < numOfElements; i++) {
                     tensorFA.set(i, tensorHFA.get(i).getFloat32());
                 }
                 yield new FP32TornadoTensor(tensorFA);
@@ -172,13 +168,15 @@ public static TornadoTensor loadTornadoTensorAsFP32(GGMLTensorEntry entry) {
                 Q8_0TornadoTensor tensorQ8_0 = Q8_0TornadoTensor.create(entry);
                 int numOfElements = tensorQ8_0.getSize();
                 FloatArray tensorFA = new FloatArray(numOfElements);
-                for(int i = 0; i < numOfElements; i++) {
+                for (int i = 0; i < numOfElements; i++) {
                     tensorFA.set(i, tensorQ8_0.getFloat(i));
                 }
                 yield new FP32TornadoTensor(tensorFA);
 
             }
-            default -> { throw new UnsupportedOperationException("Unsupported tensor type: " + tensor.type()); }
+            default -> {
+                throw new UnsupportedOperationException("Unsupported tensor type: " + tensor.type());
+            }
         };
     }
 

src/main/java/org/beehive/gpullama3/model/loader/Phi3ModelLoader.java

@@ -114,9 +114,9 @@ protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntri
 
     @Override
     protected Weights createTornadoVMWeights(Map<String, GGMLTensorEntry> tensorEntries, Phi3Configuration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                             GGMLTensorEntry outputWeight) {
         GGMLType ggmlType = outputWeight.ggmlType();
-        
+
         if (TornadoVMMasterPlan.ENABLE_TORNADOVM_INIT_TIME) {
             System.out.println("Loading model weights in TornadoVM format (loading " + ggmlType + ")");
         }

src/main/java/org/beehive/gpullama3/model/loader/Qwen2ModelLoader.java

@@ -86,7 +86,7 @@ protected Qwen2 createModel(Qwen2Configuration config, Tokenizer tokenizer, Weig
 
     @Override
     protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntries, Qwen2Configuration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                            GGMLTensorEntry outputWeight) {
 
         final int nl = config.numberOfLayers();
 
@@ -114,7 +114,7 @@ protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntri
 
     @Override
     protected Weights createTornadoVMWeights(Map<String, GGMLTensorEntry> tensorEntries, Qwen2Configuration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                             GGMLTensorEntry outputWeight) {
         GGMLType ggmlType = outputWeight.ggmlType();
 
         if (TornadoVMMasterPlan.ENABLE_TORNADOVM_INIT_TIME) {

src/main/java/org/beehive/gpullama3/model/loader/Qwen3ModelLoader.java

@@ -88,7 +88,7 @@ protected Qwen3 createModel(Qwen3Configuration config, Tokenizer tokenizer, Weig
 
     @Override
     protected Weights createStandardWeights(Map<String, GGMLTensorEntry> tensorEntries, Qwen3Configuration config, Pair<float[], float[]> ropeFreqs, GGMLTensorEntry tokenEmbeddings,
-            GGMLTensorEntry outputWeight) {
+                                            GGMLTensorEntry outputWeight) {
         float[] ropeFreqsReal = ropeFreqs.first();
         float[] ropeFreqsImag = ropeFreqs.second();
 

src/main/java/org/beehive/gpullama3/tensor/GGUF.java

@@ -96,13 +96,11 @@ public static GGUF loadGGUFMetadata(Path modelPath) throws IOException {
      * Loads tensor data from a given file channel based on the tensor metadata information.
      * The mapping is read-only and creates standard memory segments for each tensor.
      *
-     * @param fileChannel       the channel from which tensor storage is read
-     * @param tensorDataOffset  the absolute byte offset of the GGUF tensor-data section
-     * @param tensorInfos       metadata describing all GGUF tensors
-     *
+     * @param fileChannel      the channel from which tensor storage is read
+     * @param tensorDataOffset the absolute byte offset of the GGUF tensor-data section
+     * @param tensorInfos      metadata describing all GGUF tensors
      * @return a map from tensor name to {@link GGMLTensorEntry} containing
-     *         standard memory segments for each tensor
-     *
+     * standard memory segments for each tensor
      * @throws IOException if memory mapping fails or the channel cannot be read
      */
     public static Map<String, GGMLTensorEntry> loadTensorsStandard(FileChannel fileChannel, long tensorDataOffset, Map<String, GGUFTensorInfo> tensorInfos) throws IOException {
@@ -152,14 +150,11 @@ public static Map<String, GGMLTensorEntry> loadTensorsStandard(FileChannel fileC
      * before the actual tensor position, providing a writable header region
      * without modifying the underlying GGUF file.</p>
      *
-     *
-     * @param fileChannel       the channel from which tensor storage is read
-     * @param tensorDataOffset  the absolute byte offset of the GGUF tensor-data section
-     * @param tensorInfos        metadata describing all GGUF tensors
-     *
+     * @param fileChannel      the channel from which tensor storage is read
+     * @param tensorDataOffset the absolute byte offset of the GGUF tensor-data section
+     * @param tensorInfos      metadata describing all GGUF tensors
      * @return a map from tensor name to {@link GGMLTensorEntry} containing
-     *         TornadoVM-compatible memory segments for each tensor
-     *
+     * TornadoVM-compatible memory segments for each tensor
      * @throws IOException if memory mapping fails or the channel cannot be read
      */
     public static Map<String, GGMLTensorEntry> loadTensorsTornado(FileChannel fileChannel, long tensorDataOffset, Map<String, GGUFTensorInfo> tensorInfos) throws IOException {

src/main/java/org/beehive/gpullama3/tensor/tornado/TornadoTensor.java

@@ -15,6 +15,7 @@ public abstract class TornadoTensor {
 
     /**
      * Get as FloatArray (for F32 tensors).
+     *
      * @throws UnsupportedOperationException if not F32
      */
     public FloatArray asFloatArray() {
@@ -23,6 +24,7 @@ public FloatArray asFloatArray() {
 
     /**
      * Get as HalfFloatArray (for F16 tensors).
+     *
      * @throws UnsupportedOperationException if not F16
      */
     public HalfFloatArray asHalfFloatArray() {
@@ -31,6 +33,7 @@ public HalfFloatArray asHalfFloatArray() {
 
     /**
      * Get quantized scales (for Q8_0 tensors).
+     *
      * @throws UnsupportedOperationException if not quantized
      */
     public HalfFloatArray getScales() {
@@ -39,6 +42,7 @@ public HalfFloatArray getScales() {
 
     /**
      * Get quantized values (for Q8_0 tensors).
+     *
      * @throws UnsupportedOperationException if not quantized
      */
     public Int8Array getQuants() {