Adjusting the tau value issue.

Author: benedictchen

README.md

@@ -261,7 +261,7 @@ If this toolkit helps your research, please cite:
   author={Chen, Benedict},
   year={2025},
   url={https://github.com/benedictchen/meta-learning-toolkit},
-  version={2.0.0}
+  version={2.3.0}
 }
 ```
 

meta_learning/core/math_utils.py

@@ -65,15 +65,15 @@ def cosine_logits(a: torch.Tensor, b: torch.Tensor, tau: float = 10.0) -> torch.
     Args:
         a: [N, D] query features
         b: [M, D] support features  
-        tau: Temperature scaling parameter (higher = more confident)
+        tau: Temperature scaling parameter (higher = less confident)
         
     Returns:
         [N, M] cosine similarity logits
         
     Mathematical Foundation:
         - Epsilon guard prevents division by zero when ||x|| = 0
         - Temperature scaling controls prediction confidence
-        - Higher tau → sharper probability distributions
+        - Higher tau → softer probability distributions (higher entropy)
     """
     eps = _eps_like(a)
     # L2 normalize with epsilon guard against zero norms