Refactor improve codes for ranker (#8936)

### What problem does this PR solve?
Use the normalize method directly

### Type of change

- [x] Refactoring

rag/llm/rerank_model.py

 from api.utils.log_utils import log_exception
 from rag.utils import num_tokens_from_string, truncate
 
-
-def sigmoid(x):
-    return 1 / (1 + np.exp(-x))
-
-
 class Base(ABC):
     def __init__(self, key, model_name):
         pass
 
     def _compute_batch_scores(self, batch_pairs, max_length=None):
         if max_length is None:
-            scores = self._model.compute_score(batch_pairs)
+            scores = self._model.compute_score(batch_pairs, normalize=True)
         else:
-            scores = self._model.compute_score(batch_pairs, max_length=max_length)
-        scores = sigmoid(np.array(scores))
+            scores = self._model.compute_score(batch_pairs, max_length=max_length, normalize=True)
         if not isinstance(scores, Iterable):
             scores = [scores]
         return scores