当 用户的输入 匹配到 敏感词的时候，处理

dbgpt/app/apps/utils/filter.py

@@ -68,6 +68,7 @@ class DFAFilter():
         message = message.lower()
         ret = []
         start = 0
+        matched_sensitives = []  # List to store matched sensitive words
         while start < len(message):
             level = self.keyword_chains
             step_ins = 0
@@ -77,6 +78,8 @@ class DFAFilter():
                     if self.delimit not in level[char]:
                         level = level[char]
                     else:
+                        # Store the matched sensitive word
+                        matched_sensitives.append(message[start:start + step_ins])
                         ret.append(repl * step_ins)
                         start += step_ins - 1
                         is_sensitive = True
@@ -88,8 +91,8 @@ class DFAFilter():
                 ret.append(message[start])
             start += 1
 
-        #return 返回两个参数
-        return ''.join(ret), is_sensitive
+        # return 返回三个参数
+        return ''.join(ret), is_sensitive, matched_sensitives
 
 mydfafiter = DFAFilter()
 

dbgpt/app/openapi/api_v1/api_v1.py

@@ -363,17 +363,17 @@ async def chat_completions(
     await mydfafiter.parse_from_db(auth.db)
 
     # 先判断敏感词
-    dfa_result, is_sensitive = mydfafiter.filter(dialogue.user_input, "*")
-    print(dfa_result)
+    dfa_result, is_sensitive, matched_sensitives = mydfafiter.filter(dialogue.user_input, "*")
+    print(f"--->敏感词替换后的问题:{dfa_result}")
+    print(f"--->匹配到的敏感词列表:{matched_sensitives}")
 
     if is_sensitive:
         print('用户输入有敏感词')
-        result = {'code': 200, 'message': 'success',
-                  'data': [{'type': 3, 'word_name': '111', 'is_sensitive': 1, 'user_input': dfa_result}]}
+        result = f"非常抱歉，您的提问中包含了部分敏感词汇:{matched_sensitives}，我作为AI，在回答时需遵守相关法律法规和政策，我们系统对特定敏感内容进行了限制。请您尝试调整或重新组织您的问题，避开这些敏感点，我将非常乐意为您提供准确且有帮助的回答。"
         return StreamingResponse(
-            'error:1001',
+            stream_generator_sensitive(dialogue.incremental, result),
             headers=headers,
-            media_type="text/event-stream",
+            media_type="text/plain",
         )
 
     domain_type = _parse_domain_type(dialogue)
@@ -472,6 +472,33 @@ async def no_stream_generator(chat):
         msg = await chat.nostream_call()
         yield f"data: {msg}\n\n"
 
+async def stream_call_sensitive_run_one(self):
+    """
+    只运行一次
+    判断出用户输入有敏感词的时候,不需要调用大模型
+    """
+    last_output = None
+    yield last_output
+
+async def stream_generator_sensitive(
+        incremental: bool,
+        dfa_result: str,
+):
+    """
+    Generate streaming responses
+    当用户输入有敏感词的时候,统一回答
+    """
+    span = root_tracer.start_span("stream_generator_sensitive")
+    async for _ in stream_call_sensitive_run_one(1):
+        if incremental:
+            yield f"data: {dfa_result}\n\n"
+        else:
+            # TODO generate an openai-compatible streaming responses
+            yield f"data: {dfa_result}\n\n"
+        await asyncio.sleep(0.02)
+    if incremental:
+        yield "data: [DONE]\n\n"
+    span.end()
 
 async def stream_generator(chat, incremental: bool, model_name: str):
     """Generate streaming responses