minor cleanups

Author: rziga

JavaPostevanka/GPTMul/GPTMul.java

@@ -15,7 +15,7 @@ public GPTMul(Module model, Tokenizer tokenizer) {
     }
 
     public int mul(int a, int b) {
-        String prompt1 = String.format("%d*%d=0", a, b);
+        String prompt1 = String.format("%d*%d=", a, b);
         Matrix input1 = tokenizer.transform(prompt1);
         Matrix output1 = model.forward(new Matrix[] {input1})[0];
         int new_token1 = output1.rowArgMax()[3];
@@ -29,6 +29,5 @@ public int mul(int a, int b) {
         String out = tokenizer.transformInverse(pred);
         return Integer.parseInt(out);
     }
-
-
+    
 }

JavaPostevanka/Matrix/Matrix.java

@@ -309,15 +309,15 @@ public Matrix mul(Matrix other) {
     }
 
     public Matrix mul(float other) {
-        return mul(new Matrix(new float[][] {{other}}));
+        return applyUnary((x) -> x * other);
     }
 
     public Matrix div(Matrix other) {
-        return applyBinary((x, y) -> x/y, other);
+        return applyBinary((x, y) -> x / y, other);
     }
 
     public Matrix div(float other) {
-        return div(new Matrix(new float[][] {{other}}));
+        return applyUnary((x) -> x / other);
     }
 
     public Matrix reciprocal(float other) {

JavaPostevanka/NN/Layer/SoftMax.java

@@ -10,7 +10,7 @@ public class SoftMax extends Module {
     @Override
     public Matrix[] forward(Matrix[] inputs) {
         Matrix X = inputs[0];
-        Matrix expX = (X.add(X.rowMax().mul(-1))).exp();
+        Matrix expX = X.sub(X.rowMax()).exp();
         activation = expX.div(expX.rowSum());
         return new Matrix[] {activation};
     }

JavaPostevanka/NN/Optim/SGD.java

@@ -18,4 +18,5 @@ public void step() {
             p.data = p.data.sub(p.grad.mul(lr));
         }
     }
+    
 }

JavaPostevanka/Postevanka.java

@@ -10,6 +10,7 @@
 import JavaPostevanka.Trainer.Trainer;
 
 public class Postevanka {
+    
     public static void main(String[] args) {
         Random rng = new Random(1337);