q_a优化，提示词优化

service/cus_vanna_srevice.py

@@ -182,13 +182,24 @@ class OpenAICompatibleLLM(VannaBase):
                 stop=None,
                 temperature=self.temperature,
             )
+            # data = {
+            #     "model": self.model,
+            #     "prompt": prompt,
+            #     "max_tokens": self.max_tokens,
+            #     "temperature": self.temperature,
+            # }
+            # response = requests.post(
+            #     url=f"{self.api_base}/completions",
+            #     headers=self.headers,
+            #     json=data
+            # )
         else:
             if num_tokens > 3500:
                 model = "kimi"
             else:
                 model = "doubao"
 
-            print(f"Using model {model} for {num_tokens} tokens (approx)")
+            print(f"5.Using model {model} for {num_tokens} tokens (approx)")
 
             response = self.client.chat.completions.create(
                 model=model,
@@ -231,6 +242,7 @@ class OpenAICompatibleLLM(VannaBase):
             logger.info(f"sys_temp:{sys_temp}")
             llm_response = self.submit_prompt(
                 [{'role': 'system', 'content': sys_temp}, {'role': 'user', 'content': user_temp}], **kwargs)
+            llm_response = str(llm_response.strip())
             logger.info(f"llm_response:{llm_response}")
             result = {"resp": orjson.loads(extract_nested_json(llm_response))}
             logger.info(f"llm_response:{llm_response}")

template.yaml

@@ -52,10 +52,12 @@ template:
           <rule-detail>若数据库引擎是 PostgreSQL, Oracle, ClickHouse, 达梦数据库, AWS Redshift, Elasticsearch，则schema、表名、字段名、别名使用双引号，如 "schema_name"."table_name"。</rule-detail>
           <rule-detail>若数据库引擎是 MySQL, Doris，则表名、字段名、别名使用反引号，如 `table_name`。</rule-detail>
           <rule-detail>生成的SQL必须避免与数据库关键字冲突。</rule-detail>
+          <rule-detail>注意列名定义和使用的先后顺序，例如：SELECT阶段定义了列名，如果GROUP BY阶段先与SELECT阶段执行时，是不许在GROUP BY阶段引用列名的。</rule-detail>
         </rule>
         <rule>
           <rule-title>数据查询与排序</rule-title>
           <rule-detail>若未明确指定查询字段，涉及人员信息时，默认返回相关性最强的前10个字段。</rule-detail>
+          <rule-detail>当涉及部门表org_orgs的查询时注意enable启用状态和dr删除标志，且必须限制code字段值包含'CYJ'。</rule-detail>
           <rule-detail>若查询字段为 VARCHAR 或 TEXT 类型但需要计算，必须先进行合理的类型转换（如 CAST(... AS NUMERIC)）。</rule-detail>
           <rule-detail>若查询包含日期/时间字段：
             - **默认行为**：若提问未指定排序，**默认按时间字段降序排序**（即最新数据在前）。
@@ -271,22 +273,32 @@ template:
         <chat-examples>
           <example>
             <input>
-              <sql>SELECT `u`.`email` AS `email`, `u`.`id` AS `id`, `u`.`account` AS `account`, `u`.`enable` AS `enable`, `u`.`create_time` AS `create_time`, `u`.`language` AS `language`, `u`.`default_oid` AS `default_oid`, `u`.`name` AS `name`, `u`.`phone` AS `phone`, FROM `per_user` `u` LIMIT 1000</sql>
+              <sql>SELECT `u`.`email` AS `邮箱`, `u`.`id` AS `ID`, `u`.`account` AS `账号`, `u`.`enable` AS `启用状态`, `u`.`create_time` AS `创建时间`, `u`.`language` AS `语言`, `u`.`default_oid` AS `所属组织id`, `u`.`name` AS `姓名`, `u`.`phone` AS `电话`, FROM `per_user` `u` LIMIT 1000</sql>
               <user-question>查询所有用户信息</user-question>
               <chart-type></chart-type>
             </input>
             <output>
-              {{"type":"table","title":"所有用户信息","columns":[{{"name":"邮箱","value":"email"}},{{"name":"ID","value":"id"}},{{"name":"账号","value":"account"}},{{"name":"启用状态","value":"enable"}},{{"name":"创建时间","value":"create_time"}},{{"name":"语言","value":"language"}},{{"name":"所属组织ID","value":"default_oid"}},{{"name":"姓名","value":"name"}},{{"name":"Phone","value":"phone"}}]}}
+              {{"type":"table","title":"所有用户信息","columns":[{{"name":"邮箱","value":"email"}},{{"name":"ID","value":"id"}},{{"name":"账号","value":"account"}},{{"name":"启用状态","value":"enable"}},{{"name":"创建时间","value":"create_time"}},{{"name":"语言","value":"language"}},{{"name":"所属组织id","value":"default_oid"}},{{"name":"姓名","value":"name"}},{{"name":"电话","value":"phone"}}]}}
             </output>
           </example>
           <example>
             <input>
-              <sql>SELECT `o`.`name` AS `org_name`, COUNT(`u`.`id`) AS `user_count` FROM `per_user` `u` JOIN `per_org` `o` ON `u`.`default_oid` = `o`.`id` GROUP BY `o`.`name` ORDER BY `user_count` DESC LIMIT 1000</sql>
+              <sql>SELECT `o`.`name` AS `org_name`, COUNT(`u`.`id`) AS `人数` FROM `per_user` `u` JOIN `per_org` `o` ON `u`.`default_oid` = `o`.`id` GROUP BY `o`.`name` ORDER BY `user_count` DESC LIMIT 1000</sql>
               <user-question>饼图展示各个组织的人员数量</user-question>
               <chart-type> pie </chart-type>
             </input>
             <output>
-              {{"type":"pie","title":"组织人数统计","axis":{{"y":{{"name":"人数","value":"user_count"}},"series":{{"name":"组织名称","value":"org_name"}}}}}}
+              {{"type":"pie","title":"组织人数统计","axis":{{"y":{{"name":"人数","value":"人数"}},"series":{{"name":"org_name","value":"org_name"}}}}}}
+            </output>
+          </example>
+          <example>
+            <input>
+              <sql>SELECT COUNT(*) AS "总人数", SUM(CASE WHEN p."gender" = '1' THEN 1 ELSE 0 END) AS "男员工数", SUM(CASE WHEN p."gender" = '2' THEN 1 ELSE 0 END) AS "女员工数" FROM "YJOA_APPSERVICE_DB"."t_pr3rl2oj_yj_person_database" p WHERE p."dr" = 0 ORDER BY "总人数" DESC LIMIT 1000;</sql>
+              <user-question>表格展示不同性别人员数量，及人员总数</user-question>
+              <chart-type> bar </chart-type>
+            </input>
+            <output>
+              {{'type': 'table', 'title': '员工性别统计', 'columns': [{{"name":"总人数","value":"total_persons"}},{{"name":"男员工数","value":"male_count"}},{{"name":"女员工数","value":"female_count"}}]}}
             </output>
           </example>
         </chat-examples>

util/q_and_a_dict.py

@@ -8,7 +8,7 @@ question_and_answer = [
                FROM YJOA_APPSERVICE_DB.t_pr3rl2oj_yj_person_database p
                WHERE p.internal_dept IN (SELECT "id"
                                          FROM "IUAP_APDOC_BASEDOC"."org_orgs" START
-               WITH "name" LIKE '%综合处%'
+               WITH "name"||"shortname" LIKE '%综合处%'
                CONNECT BY PRIOR "id" = "parentid"
                    )
                       AND p.dr = 0
@@ -198,7 +198,7 @@ question_and_answer = [
                   from YJOA_APPSERVICE_DB.t_pr3rl2oj_yj_person_database
                   where internal_dept in (SELECT "id"
                                           FROM "IUAP_APDOC_BASEDOC"."org_orgs" START
-                  WITH "name" LIKE '%数信中心%'
+                  WITH "name"||"shortname" LIKE '%数信中心%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
                   CONNECT BY PRIOR "id" = "parentid"
                       )
 
@@ -224,7 +224,7 @@ question_and_answer = [
                     JOIN IUAP_APDOC_BASEDOC.org_orgs o ON p.internal_dept = o.id
                     WHERE p.internal_dept IN (
                         SELECT "id" FROM "IUAP_APDOC_BASEDOC"."org_orgs" 
-                        START WITH "name" LIKE '%数信中心%'
+                        START WITH "name"||"shortname" LIKE '%数信中心%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
                         CONNECT BY PRIOR "id" = "parentid"
                     )
                     AND p.dr = 0
@@ -262,7 +262,7 @@ question_and_answer = [
                     AND ps."date_value" LIKE '2025-10%'
                     and p.internal_dept in (SELECT "id"
                                           FROM "IUAP_APDOC_BASEDOC"."org_orgs" START
-                  WITH "name" LIKE '%数信中心%'
+                  WITH "name" || "shortname" LIKE '%数信中心%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
                   CONNECT BY PRIOR "id" = "parentid"
                       )
                     AND p."dr" = 0
@@ -279,7 +279,7 @@ question_and_answer = [
                   '''
     },
     {
-        "question": "员工年龄段分布图",
+        "question": "数信中心员工年龄段分布图",
         "answer": '''
                     SELECT 
                         CASE 
@@ -301,7 +301,10 @@ question_and_answer = [
                             WHERE "parentid" IN (
                                 SELECT "id" 
                                 FROM "IUAP_APDOC_BASEDOC"."org_orgs" 
-                            )
+                                WHERE name || shortname LIKE '%数信中心%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
+                            ) OR id IN (SELECT "id" 
+                                FROM "IUAP_APDOC_BASEDOC"."org_orgs" 
+                                WHERE name || shortname LIKE '%数信中心%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%')
                         )
                     GROUP BY 
                         CASE 
@@ -328,14 +331,15 @@ question_and_answer = [
                   WHERE p."dr" = 0
                     AND o."id" IN (SELECT "id"
                                    FROM "IUAP_APDOC_BASEDOC"."org_orgs" START
-                  WITH "name" LIKE '%综合处%'
+                  WITH "name"||"shortname" LIKE '%综合处%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
                   CONNECT BY PRIOR "id" = "parentid")
                          AND p."birthday" IS NOT NULL
                          AND p."birthday" != ''
                   ORDER BY o."name" ASC, p."birthday" DESC
                       LIMIT 1000;
                   '''
-    },{
+    },
+    {
         "question": "查询综合处下面的员工的平均年龄",
         "answer": '''
                   SELECT ROUND(AVG(2025 - CAST(SUBSTR(p."birthday", 1, 4) AS INT)), 2) AS "平均年龄"
@@ -344,12 +348,45 @@ question_and_answer = [
                   WHERE p."dr" = 0
                     AND o."id" IN (SELECT "id"
                                    FROM "IUAP_APDOC_BASEDOC"."org_orgs" START
-                  WITH "name" LIKE '%综合处%'
+                  WITH "name"||"shortname" LIKE '%综合处%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
                   CONNECT BY PRIOR "id" = "parentid")
                          AND p."birthday" IS NOT NULL
                          AND p."birthday" != ''
                       LIMIT 1000;
                   '''
+    },
+    {
+        "question": "综合处有几个人，男员工，女员工分别有多少",
+        "answer": '''
+                    SELECT COUNT(*) AS "总人数", 
+                    SUM(CASE WHEN p."gender" = '1' THEN 1 ELSE 0 END) AS "男员工数", 
+                    SUM(CASE WHEN p."gender" = '2' THEN 1 ELSE 0 END) AS "女员工数"
+                    FROM "YJOA_APPSERVICE_DB"."t_pr3rl2oj_yj_person_database" p
+                    WHERE p."internal_dept" IN (
+                        SELECT "id"
+                        FROM "IUAP_APDOC_BASEDOC"."org_orgs"
+                        START WITH "name" LIKE '%综合处%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
+                        CONNECT BY PRIOR "id" = "parentid"
+                    )
+                    AND p."dr" = 0
+                    LIMIT 1000;
+                  '''
+    },
+    {
+        "question": "综合处女员工有多少",
+        "answer": '''
+                    SELECT (SELECT COUNT(*) 
+                    FROM "YJOA_APPSERVICE_DB"."t_pr3rl2oj_yj_person_database" p 
+                    WHERE p."internal_dept" IN (SELECT "id"
+                               FROM "IUAP_APDOC_BASEDOC"."org_orgs"
+                               START WITH "name"||"shortname" LIKE '%综合处%' AND "dr"=0 AND "enable"=1 AND "code" LIKE '%CYJ%'
+                               CONNECT BY PRIOR "id" = "parentid"
+                              )
+                    AND p."gender" = '2'
+                    AND p."dr" = 0
+                    ) AS "女员工数"
+                      '''
     }
 
+
 ]

util/train_ddl.py

@@ -963,6 +963,35 @@ org_orgs_ddl = '''
       "role": "dimension",
       "tags": ["部门名称","部门简称","部门缩写"]
     },
+    {
+      "name": "enable",
+      "type": "INT",
+      "comment": "启用状态",
+      "value":{
+        "1":"启用"
+        "2":"未启用"
+      },
+      "role": "dimension",
+      "tags": ["部门启用","部门状态","是否启用"]
+    },
+    {
+      "name": "level",
+      "type": "INT",
+      "comment": "部门或单位层级",
+      "role": "dimension",
+      "tags": ["部门层级","级别"]
+    },
+    {
+      "name": "dr",
+      "type": "INT",
+      "comment": "删除标志",
+      "value": {
+        "0": "正常",
+        "1": "删除"
+      },
+      "role": "dimension",
+      "tags": ["状态标识", "软删除", "枚举"]
+    }
   ],
  
   "tags": ["部门id","部门信息","部门名称"]

util/utils.py

@@ -1,7 +1,8 @@
+import logging
 from typing import Optional
 
 from orjson import orjson
-
+logger = logging.getLogger(__name__)
 keywords = {
     # "gender":{"1":"男","2":"女"},
     "person_status":{"1":"草稿","2":"审批中","3":"制卡中","4":"已入库","5":"停用"},
@@ -43,7 +44,8 @@ def extract_nested_json(text):
     stack = []
     start_index = -1
     results = []
-
+    if not text:
+        logger.warning("extract_nested_json: text is empty")
     for i, char in enumerate(text):
         if char in '{[':
             if not stack:  # 记录起始位置