Update src/qa.py

src/qa.py

@@ -112,27 +112,45 @@ def retrieve_chunks(query: str, index, chunks: list, top_k: int = 5):
         return []
 
 # ==========================================================
-# 6️⃣ Answer Generation (with reasoning toggle)
+# 6️⃣ Answer Generation (Enhanced — Balanced Reasoning + Speed)
 # ==========================================================
 def generate_answer(query: str, retrieved_chunks: list, reasoning_mode: bool = False):
-    """Generate answers with or without reasoning."""
+    """
+    Generates answers with or without reasoning.
+    - reasoning_mode=True → uses world knowledge + logic (slower, more explanatory)
+    - reasoning_mode=False → sticks to chunks for fast factual accuracy
+    """
     if not retrieved_chunks:
         return "Sorry, I couldn’t find relevant information in the document."
 
+    # Build the prompt with selected mode
     context = "\n".join(chunk.strip() for chunk in retrieved_chunks)
-    prompt = (REASONING_PROMPT if reasoning_mode else STRICT_PROMPT).format(context=context, query=query)
+    prompt = (REASONING_PROMPT if reasoning_mode else STRICT_PROMPT).format(
+        context=context,
+        query=query
+    )
 
     try:
+        # ⚙️ Tuned generation settings for balance
         result = _answer_model(
             prompt,
-            max_new_tokens=150 if reasoning_mode else 120,
-            temperature=0.5 if reasoning_mode else 0.2,
-            do_sample=reasoning_mode,
+            max_new_tokens=180 if reasoning_mode else 120,   # let reasoning finish sentences
+            temperature=0.6 if reasoning_mode else 0.2,      # more creative but controlled
+            top_p=0.9 if reasoning_mode else 0.8,            # smooth probability cutoff
+            do_sample=reasoning_mode,                        # only sample when reasoning
+            early_stopping=True,
             pad_token_id=_tokenizer.eos_token_id,
         )
+
+        # Clean answer text
         answer = result[0]["generated_text"].strip()
         if "Answer:" in answer:
             answer = answer.split("Answer:")[-1].strip()
+
+        # ✅ Prevents mid-sentence cutoffs
+        if answer.endswith(("and", "or", ",")):
+            answer += " ..."
+
         return answer
 
     except Exception as e: