update: prompt

prompt_templete.py

@@ -475,16 +475,16 @@ Your task is to analyze the user's input and chat history, then output a single,
 **INSTRUCTIONS:**
 1.  **Rewrite:** Create a clear, standalone Vietnamese question based on the user's input and chat history. Use official legal terms for legal queries.
 2.  **Classify:** Categorize the rewritten question into ONE of these types:
-    - `legal_rag`: A specific question about Vietnamese law.
+    - `legal_rag`: A specific question about Vietnamese law. This includes regulations, procedures, penalties, **and factual data stipulated in legal documents (e.g., administrative units, government organization, official statistics from legal texts).**
     - `out_of_scope_legal`: A question about non-Vietnamese law.
     - `chit_chat`: Greetings, thanks, feedback, or questions about the bot.
-    - `general_knowledge`: A non-legal, general knowledge question.
+    - `general_knowledge`: A non-legal, general knowledge question whose answer is NOT found in legal documents.
     - `ambiguous_legal_topic`: A legal topic mentioned without a specific question.
 
 ---
 **EXAMPLES:**
 
-# Example 1: In-scope Legal
+# Example 1: In-scope Legal (Penalty)
 - User Input: "xe may vuot den do bi phat bao nhieu tien"
 - Output:
     {{
@@ -516,6 +516,14 @@ Your task is to analyze the user's input and chat history, then output a single,
       "classification": "legal_rag",
       "rewritten_question": "Thủ tục ly hôn theo yêu cầu của một bên (ly hôn đơn phương) được quy định như thế nào?"
     }}
+
+# Example 5: In-scope Legal (Factual Data from Legal Document)
+- User Input: "sau khi sáp nhập, việt nam có bao nhiêu tỉnh thành phố"
+- Output:
+    {{
+      "classification": "legal_rag",
+      "rewritten_question": "Số lượng đơn vị hành chính cấp tỉnh của Việt Nam sau các đợt sáp nhập gần nhất là bao nhiêu theo quy định của pháp luật?"
+    }}
 ---
 
 **Chat History:**
@@ -557,14 +565,18 @@ Your task is to answer the QUESTION based **strictly** on the provided CONTEXT.
 **ANSWER GENERATION PROCESS (Internal thought process):**
 
 **1. Analysis:**
-   - **Core Request:** [Summarize what the user wants]
+   - **Core Request:** [Summarize what the user wants. Is it a rule, a penalty, a procedure, a definition, or a specific fact/number?]
    - **Main Subject & Action:** [Identify key entities and events]
 
 **2. Context Evaluation:**
-   - [List relevant documents from CONTEXT and why. If none, state that.]
+   - [List relevant documents from CONTEXT and why. Note the source (e.g., Nghị định 100/2019/NĐ-CP, Nghị quyết 1211/2016/UBTVQH13). If none, state that.]
 
 **3. Response Plan:**
-   - [Outline the answer structure based on selected documents.]
+   - [Outline the answer structure.
+      - If it's about a rule/penalty, I will use the structure: Regulation, Penalty, etc.
+      - If it's a factual question (e.g., 'how many provinces?', 'what is the area?'), I will directly state the fact/number found in the context.
+      - If it's a definition, I will provide the definition as written in the law.
+      - The plan is to present the information clearly and concisely.]
 
 **4. Final Answer (Use this exact format):**
 
@@ -572,10 +584,14 @@ Your task is to answer the QUESTION based **strictly** on the provided CONTEXT.
 
 ### [A clear, concise title for the answer in the user's language]
 
-[Present the answer here. Use bullet points for key information.]
--   **Quy định (Regulation):** [Content of the regulation]
--   **Mức phạt (Penalty):** [Details on the penalty, if any]
--   **Biện pháp bổ sung (Additional Measures):** [Details on supplementary penalties]
+[Present the answer here.
+-   Answer directly and clearly based on the Response Plan.
+-   Use paragraphs for explanations and bullet points for lists or key details.
+-   **If the question asks for a specific number, list, or fact (e.g., 'how many provinces?', 'what is the new area of Tuyen Quang?'), provide that information directly and concisely first.**
+-   For questions about rules and penalties, you can use a structure like:
+    -   **Quy định (Regulation):** [Content of the regulation]
+    -   **Mức phạt (Penalty):** [Details on the penalty, if any]
+    -   **Biện pháp bổ sung (Additional Measures):** [Details on supplementary penalties]]
 
 ### Nguồn tham khảo (Source)