Update model_utils.py

model_utils.py

@@ -31,7 +31,7 @@ model.eval()
 # -----------------------------
 EMBED_MODEL_NAME = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
 embed_model = SentenceTransformer(EMBED_MODEL_NAME)
-# move embedding model to same device (optional but faster on GPU)
+# (optional) move embedding model to same device; OK to leave on CPU if you want
 embed_model = embed_model.to(device)
 
 
@@ -72,7 +72,7 @@ SYSTEM_PROMPT = (
     "ສໍາລັບນັກຮຽນຊັ້ນ ມ.1. "
     "ຕອບແຕ່ພາສາລາວ ໃຫ້ຕອບສັ້ນໆ 2–3 ປະໂຫຍກ ແລະເຂົ້າໃຈງ່າຍ. "
     "ໃຫ້ອີງຈາກຂໍ້ມູນຂ້າງລຸ່ມນີ້ເທົ່ານັ້ນ. "
-    "ຖ້າຂໍ້ມູນບໍ່ພຽງພໍ ຫຼືບຍັງບໍ່ຊັດເຈນ ໃຫ້ບອກວ່າບໍ່ແນ່ໃຈ."
+    "ຖ້າຂໍ້ມູນບໍ່ພຽງພໍ ຫຼືບໍ່ຊັດເຈນ ໃຫ້ບອກວ່າບໍ່ແນ່ໃຈ."
 )
 
 
@@ -105,7 +105,7 @@ def retrieve_context(question: str, max_entries: int = MAX_CONTEXT_ENTRIES) -> s
         top_entries = [qa_store.ENTRIES[i] for i in top_indices.tolist()]
 
     # Build context string for the prompt
-    context_blocks = []
+    context_blocks: List[str] = []
     for e in top_entries:
         header = (
             f"[ຊັ້ນ {e.get('grade','')}, "
@@ -200,8 +200,8 @@ def answer_from_qa(question: str) -> Optional[str]:
     best_answer: Optional[str] = None
 
     for item in qa_store.ALL_QA_KNOWLEDGE:
-        stored_terms = [t for t in item["norm_q"].split(" ") if len(t) > 1]:
-            overlap = sum(1 for t in q_terms if t in stored_terms)
+        stored_terms = [t for t in item["norm_q"].split(" ") if len(t) > 1]
+        overlap = sum(1 for t in q_terms if t in stored_terms)
         if overlap > best_score:
             best_score = overlap
             best_answer = item["a"]