Modified the squeeze() application to embeddings

handler.py

@@ -115,18 +115,19 @@ class EndpointHandler:
         # tensor = torch.stack(frame_embedding)
 
         # detach text emb from graph, move to CPU, and convert to numpy array
-        self.logger.info("Squeezing tensor")
-        batch_emb = frame_embedding.squeeze(0)
+        # self.logger.info("Squeezing tensor")
+        # batch_emb = frame_embedding.squeeze(0)
 
         # Check the shape of the tensor
-        self.logger.info(f"Shape of the batch_emb tensor: {batch_emb.shape}")
+        self.logger.info(f"Shape of the batch_emb tensor: {frame_embedding.shape}")
 
         # Normalize the embeddings if it's a 2D tensor
-        if batch_emb.dim() == 2:
+        if frame_embedding.dim() == 2:
             self.logger.info("Normalizing embeddings")
-            batch_emb = torch.nn.functional.normalize(batch_emb, p=2, dim=1)
+            batch_emb = torch.nn.functional.normalize(frame_embedding, p=2, dim=1)
         else:
             self.logger.info("Skipping normalization due to tensor shape")
+            batch_emb = frame_embedding.squeeze(0)
 
         self.logger.info("Converting into numpy array")
         batch_emb = batch_emb.cpu().detach().numpy()