Refactor tensor creation methods to use static factory methods and remove redundant constructors.

Author: orionpapadakis

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

@@ -126,8 +126,8 @@ public static TornadoTensor loadTornadoTensor(GGMLTensorEntry entry) {
         GGMLType ggmlType = entry.ggmlType();
         int size = FloatTensor.numberOfElements(entry.shape());
         return switch (ggmlType) {
-            case F32 -> new FP32TornadoTensor(size, entry.memorySegment());
-            case F16 -> new FP16TornadoTensor(size, entry.memorySegment());
+            case F32 -> FP32TornadoTensor.fromTornadoMemorySegment(entry.memorySegment());
+            case F16 -> FP16TornadoTensor.fromTornadoMemorySegment(entry.memorySegment());
             case Q8_0 -> Q8_0TornadoTensor.create(entry);
             case Q4_0 -> throw new UnsupportedOperationException("Q4 format not supported yet");
             default -> throw new UnsupportedOperationException("Quantization format " + ggmlType);

src/main/java/org/beehive/gpullama3/tensor/tornado/FP16TornadoTensor.java

@@ -13,8 +13,8 @@ public FP16TornadoTensor(HalfFloatArray halfFloatArray) {
         this.tornadoNativeArray = halfFloatArray;
     }
 
-    public FP16TornadoTensor(MemorySegment segment) {
-        this.tornadoNativeArray = new HalfFloatArray(segment);
+    public static FP16TornadoTensor fromTornadoMemorySegment(MemorySegment segment) {
+        return new FP16TornadoTensor(HalfFloatArray.fromSegmentShallow(segment));
     }
 
     @Override

src/main/java/org/beehive/gpullama3/tensor/tornado/FP32TornadoTensor.java

@@ -12,8 +12,8 @@ public FP32TornadoTensor(FloatArray floatArray) {
         this.tornadoNativeArray = floatArray;
     }
 
-    public FP32TornadoTensor(MemorySegment segment) {
-        this.tornadoNativeArray = new FloatArray(segment);
+    public static FP32TornadoTensor fromTornadoMemorySegment(MemorySegment segment) {
+        return new FP32TornadoTensor(FloatArray.fromSegmentShallow(segment));
     }
 
     @Override

src/main/java/org/beehive/gpullama3/tensor/tornado/Q8_0TornadoTensor.java

@@ -41,11 +41,6 @@ public Int8Array getQuants() {
         return quants;
     }
 
-    @Override
-    public int size() {
-        return size;
-    }
-
     @Override
     public GGMLType type() {
         return GGMLType.Q8_0;
@@ -62,7 +57,7 @@ public MemorySegment asMemorySegment() {
      * @return Dequantized float value
      */
     public float getFloat(int index) {
-        assert 0 <= index && index < size;
+        assert 0 <= index;
         int blockIdx = index / GGMLType.Q8_0.getBlockSize();
         float scale = scales.get(blockIdx).getFloat32();
         byte quant = quants.get(index);
@@ -108,6 +103,6 @@ public static Q8_0TornadoTensor create(GGMLTensorEntry entry) {
             }
         }
 
-        return new Q8_0TornadoTensor(size, scales, quants, q8Segment);
+        return new Q8_0TornadoTensor(scales, quants, q8Segment);
     }
 }