Update src/qa.py

src/qa.py

@@ -112,28 +112,45 @@ def retrieve_chunks(query: str, index, chunks: list, top_k: int = 5):
 # ==========================================================
 def generate_answer(query: str, retrieved_chunks: list, reasoning_mode: bool = False):
     """
-    Generate answers with or without reasoning.
-    reasoning_mode=True → reasoning-rich, slightly slower
-    reasoning_mode=False → strict factual, fast
+    Generate answers with Phi-2.
+    - reasoning_mode=False → strict factual, fast
+    - reasoning_mode=True  → analytical, richer reasoning (slower)
     """
     if not retrieved_chunks:
         return "Sorry, I couldn’t find relevant information in the document."
 
     context = "\n".join(chunk.strip() for chunk in retrieved_chunks)
+
+    # 🧠 Reasoning prompt: encourages explanation, not just lookup
+    REASONING_PROMPT = (
+        "You are an expert assistant with strong reasoning skills.\n"
+        "Think step by step and form a detailed, logical answer.\n"
+        "You can combine hints from the context with your general understanding.\n"
+        "If the context doesn't mention the answer, acknowledge that.\n\n"
+        "Context:\n{context}\n\nQuestion: {query}\n\nAnswer:"
+    )
+
+    # ⚡ Strict factual prompt
+    STRICT_PROMPT = (
+        "Answer based ONLY on the context below.\n"
+        "If the answer isn’t in the context, say: 'I don't know based on the provided document.'\n\n"
+        "Context:\n{context}\n\nQuestion: {query}\nAnswer:"
+    )
+
     prompt = (REASONING_PROMPT if reasoning_mode else STRICT_PROMPT).format(context=context, query=query)
 
     try:
         if reasoning_mode:
-            # 🧠 “Brainy mode” — restores the earlier expressive logic
+            # 🧩 The “brainy” config that produced the great long answer
             result = _answer_model(
                 prompt,
                 max_new_tokens=180,
-                temperature=0.4,
-                do_sample=False,  # deterministic reasoning chain
+                temperature=0.45,
+                do_sample=False,  # reasoning but deterministic
                 pad_token_id=_tokenizer.eos_token_id,
             )
         else:
-            # ⚡ Strict factual fast mode
+            # ⚡ Fast factual config
             result = _answer_model(
                 prompt,
                 max_new_tokens=120,
@@ -142,10 +159,10 @@ def generate_answer(query: str, retrieved_chunks: list, reasoning_mode: bool = F
                 pad_token_id=_tokenizer.eos_token_id,
             )
 
-        raw = result[0]["generated_text"]
+        raw = result[0]["generated_text"].strip()
         if "Answer:" in raw:
             raw = raw.split("Answer:")[-1].strip()
-        return raw.strip()
+        return raw
 
     except Exception as e:
         print(f"⚠️ Generation failed: {e}")