Initialize label_attention_matrix to None before text_embedder branch

Co-authored-by: meilame-tayebjee <114609737+meilame-tayebjee@users.noreply.github.com>

Author: Copilot

torchTextClassifiers/model/model.py

@@ -134,6 +134,7 @@ def forward(
                 Raw, not softmaxed.
         """
         encoded_text = input_ids  # clearer name
+        label_attention_matrix = None
         if self.text_embedder is None:
             x_text = encoded_text.float()
         else: