Update cross_encoder_reranking_train.py

cross_encoder_reranking_train.py

@@ -17,9 +17,13 @@ device = 'cuda' if torch.cuda.is_available() else 'cpu'
 embedder = SentenceTransformer("intfloat/e5-large-v2").to(device)
 
 
+# def embed_text_list(texts):
+#     return embedder.encode(texts, convert_to_tensor=False, device=device)
+
 def embed_text_list(texts):
-    # return embedder.encode(texts, convert_to_tensor=False, device=device)
-    return embedder.encode([f"query: {texts}"], convert_to_tensor=False, device=device)
+    # E5 models expect "query: " prefix for proper embedding behavior
+    formatted_texts = [f"query: {text}" for text in texts]
+    return embedder.encode(formatted_texts, convert_to_tensor=False, device=device)
 
 def rank_by_centrality(texts):
     embeddings = embed_text_list(texts)