Update smooth_l1_loss in losses.py (#1974)

According the definition of smooth_l1_loss, the line 

diff = predictions - targets

Should be updated to 

diff = mx.abs(predictions - targets)

After the modification, the result is consistent with PyTorch smooth_l1_loss

python/mlx/nn/losses.py

@@ -373,7 +373,7 @@ def smooth_l1_loss(
             f"targets shape {targets.shape}."
         )
 
-    diff = predictions - targets
+    diff = mx.abs(predictions - targets)
     loss = mx.where(
         diff < beta, 0.5 * mx.square(diff) / beta, mx.abs(diff) - 0.5 * beta
     )