minor changes.

app.py

@@ -37,31 +37,29 @@ def retrieve_top_k(query, k=5):
     return results
 
 def build_prompt(query, retrieved_docs):
-    context_text = "\n".join([
-        f"- {doc['text']}" for _, doc in retrieved_docs.iterrows()
-    ])
+    context = "\n".join([f"- {row['text']}" for _, row in retrieved_docs.iterrows()])
+    prompt = f"""Instruction: Based on the following patient symptoms and clinical notes, provide a detailed diagnostic reasoning or suggestion.
 
-    prompt = f"""You are a medical assistant trained on clinical reasoning data.
-Given the following patient query and related clinical observations, generate a diagnostic explanation.
-
-Patient Query:
-{query}
+Input:
+Patient Query: {query}
 
 Clinical Context:
-{context_text}
+{context}
 
-Diagnostic Explanation:"""
+Response:"""
     return prompt
 
-def generate_local_answer(prompt, max_new_tokens=256):
-    input_ids = tokenizer(prompt, return_tensors="pt").input_ids  # CPU only
+
+def generate_local_answer(prompt, max_new_tokens=512):
+    input_ids = tokenizer(prompt, return_tensors="pt", truncation=True).input_ids  # CPU
     output = generation_model.generate(
         input_ids=input_ids,
         max_new_tokens=max_new_tokens,
-        temperature=0.7,
-        do_sample=True,
+        temperature=0.4,         # Less random
         top_k=50,
         top_p=0.95,
+        repetition_penalty=1.2,  # Encourage more varied output
+        num_beams=2              # Mild beam search for quality
     )
     decoded = tokenizer.decode(output[0], skip_special_tokens=True)
     return decoded.strip()