update

app.py

@@ -57,6 +57,7 @@ if is_env_local:
         OPEN_AI_KEY = config["OPEN_AI_KEY"]
         OPEN_AI_ASSISTANT_ID_GPT4_BOT1 = config["OPEN_AI_ASSISTANT_ID_GPT4_BOT1"]
         OPEN_AI_ASSISTANT_ID_GPT3_BOT1 = config["OPEN_AI_ASSISTANT_ID_GPT3_BOT1"]
+        OPEN_AI_MODERATION_BOT1 = config["OPEN_AI_MODERATION_BOT1"]
         OPEN_AI_KEY_BOT2 = config["OPEN_AI_KEY_BOT2"]
         OPEN_AI_ASSISTANT_ID_GPT4_BOT2 = config["OPEN_AI_ASSISTANT_ID_GPT4_BOT2"]
         OPEN_AI_ASSISTANT_ID_GPT3_BOT2 = config["OPEN_AI_ASSISTANT_ID_GPT3_BOT2"]
@@ -76,6 +77,7 @@ else:
     OPEN_AI_KEY = os.getenv("OPEN_AI_KEY")
     OPEN_AI_ASSISTANT_ID_GPT4_BOT1 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT4_BOT1")
     OPEN_AI_ASSISTANT_ID_GPT3_BOT1 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT3_BOT1")
+    OPEN_AI_MODERATION_BOT1 = os.getenv("OPEN_AI_MODERATION_BOT1")
     OPEN_AI_KEY_BOT2 = os.getenv("OPEN_AI_KEY_BOT2")
     OPEN_AI_ASSISTANT_ID_GPT4_BOT2 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT4_BOT2")
     OPEN_AI_ASSISTANT_ID_GPT3_BOT2 = os.getenv("OPEN_AI_ASSISTANT_ID_GPT3_BOT2")
@@ -122,8 +124,20 @@ def check_open_ai_access(open_ai_api_key):
         return False
 
 open_ai_api_key_assistant_id_list = [
-    {"account":"bot1", "open_ai_api_key": OPEN_AI_KEY, "assistant_gpt4_id": OPEN_AI_ASSISTANT_ID_GPT4_BOT1, "assistant_gpt3_id": OPEN_AI_ASSISTANT_ID_GPT3_BOT1},
-    {"account":"bot2", "open_ai_api_key": OPEN_AI_KEY_BOT2, "assistant_gpt4_id": OPEN_AI_ASSISTANT_ID_GPT4_BOT2, "assistant_gpt3_id": OPEN_AI_ASSISTANT_ID_GPT3_BOT2},
+    {
+        "account":"bot1", 
+        "open_ai_api_key": OPEN_AI_KEY, 
+        "assistant_gpt4_id": OPEN_AI_ASSISTANT_ID_GPT4_BOT1, 
+        "assistant_gpt3_id": OPEN_AI_ASSISTANT_ID_GPT3_BOT1,
+        "moderation": OPEN_AI_MODERATION_BOT1
+    },
+    {
+        "account":"bot2", 
+        "open_ai_api_key": OPEN_AI_KEY_BOT2, 
+        "assistant_gpt4_id": OPEN_AI_ASSISTANT_ID_GPT4_BOT2, 
+        "assistant_gpt3_id": OPEN_AI_ASSISTANT_ID_GPT3_BOT2,
+        "moderation": OPEN_AI_MODERATION_BOT1
+    },
 ]
 for open_ai_api_key_assistant_id in open_ai_api_key_assistant_id_list:
     account = open_ai_api_key_assistant_id["account"]
@@ -132,6 +146,7 @@ for open_ai_api_key_assistant_id in open_ai_api_key_assistant_id_list:
         OPEN_AI_CLIENT = OpenAI(api_key=open_ai_api_key)
         OPEN_AI_ASSISTANT_ID_GPT4 = open_ai_api_key_assistant_id["assistant_gpt4_id"]
         OPEN_AI_ASSISTANT_ID_GPT3 = open_ai_api_key_assistant_id["assistant_gpt3_id"]
+        OPEN_AI_MODERATION_CLIENT = OpenAI(api_key=open_ai_api_key_assistant_id["moderation"])
         print(f"OpenAI access is OK, account: {account}")
         break
 
@@ -2177,6 +2192,18 @@ def get_instructions(content_subject, content_grade, key_moments, socratic_mode=
     """
     return instructions
 
+def get_chat_moderation(user_content):
+    # response = client.moderations.create(input=text)
+    response = OPEN_AI_MODERATION_CLIENT.moderations.create(input=user_content)
+    response_dict = response.model_dump()
+    is_flagged = response_dict['results'][0]['flagged']
+    print("========get_chat_moderation==========")
+    print(f"is_flagged: {is_flagged}")
+    print(response_dict)
+    print("========get_chat_moderation==========")
+
+    return is_flagged, response_dict
+
 def chat_with_any_ai(ai_type, password, video_id, user_data, transcript_state, key_moments, user_message, chat_history, content_subject, content_grade, questions_answers_json, socratic_mode=False, thread_id=None, ai_name=None):
     print(f"ai_type: {ai_type}")
     print(f"user_data: {user_data}")
@@ -2194,10 +2221,14 @@ def chat_with_any_ai(ai_type, password, video_id, user_data, transcript_state, k
     
     verify_chat_limit(chat_history, CHAT_LIMIT)
 
+    is_flagged, response_dict = get_chat_moderation(user_message)
     if ai_type == "chat_completions":
-        chatbot_config = get_chatbot_config(ai_name, transcript_state, key_moments, content_subject, content_grade, video_id, socratic_mode)
-        chatbot = Chatbot(chatbot_config)
-        response_text = chatbot.chat(user_message, chat_history)
+        if is_flagged:
+            response_text = "您的留言已被標記為不當內容，請重新發送。"
+        else:
+            chatbot_config = get_chatbot_config(ai_name, transcript_state, key_moments, content_subject, content_grade, video_id, socratic_mode)
+            chatbot = Chatbot(chatbot_config)
+            response_text = chatbot.chat(user_message, chat_history)
         # if thread_id is none, create random thread_id + timestamp
         if thread_id is None or thread_id == "":
             thread_id = "thread_" + str(uuid.uuid4()) + str(int(time.time()))
@@ -2212,22 +2243,12 @@ def chat_with_any_ai(ai_type, password, video_id, user_data, transcript_state, k
             "socratic_mode": str(socratic_mode),
             "assistant_id": ai_name,
             "is_streaming": "false",
+            "moderation_is_flagged": str(is_flagged),
+            "moderation_response_dict": str(response_dict)
         }
     elif ai_type == "assistant":
         client = OPEN_AI_CLIENT
-        assistant_id = OPEN_AI_ASSISTANT_ID_GPT4 #GPT 4 turbo
-        if isinstance(key_moments, str):
-            key_moments_json = json.loads(key_moments)
-        else:
-            key_moments_json = key_moments
-        # key_moments_json remove images
-        for moment in key_moments_json:
-            moment.pop('images', None)
-            moment.pop('end', None)
-            moment.pop('transcript', None)
-        key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)
-        instructions = get_instructions(content_subject, content_grade, key_moments_text, socratic_mode)
-        print(f"=== instructions:{instructions} ===")
+        assistant_id = OPEN_AI_ASSISTANT_ID_GPT4
         metadata={
             "video_id": video_id,
             "user_data": user_data,
@@ -2236,10 +2257,28 @@ def chat_with_any_ai(ai_type, password, video_id, user_data, transcript_state, k
             "socratic_mode": str(socratic_mode),
             "assistant_id": assistant_id,
             "is_streaming": "false",
+            "moderation_is_flagged": str(is_flagged),
+            "moderation_response_dict": str(response_dict)
         }
-        user_message_note = "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，絕對不要重複 user 的問句，請用引導的方式指引方向，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，請在回答的最後標註【參考：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內，回答時如果講到數學專有名詞，請用數學符號代替文字（Latex 用 $ 字號 render, ex: $x^2$)"
-        user_content = user_message + user_message_note
-        response_text, thread_id = handle_conversation_by_open_ai_assistant(client, user_content, instructions, assistant_id, thread_id, metadata, fallback=True)
+
+        if is_flagged:
+            response_text = "您的留言已被標記為不當內容，請重新發送。"
+        else:
+            if isinstance(key_moments, str):
+                key_moments_json = json.loads(key_moments)
+            else:
+                key_moments_json = key_moments
+            # key_moments_json remove images
+            for moment in key_moments_json:
+                moment.pop('images', None)
+                moment.pop('end', None)
+                moment.pop('transcript', None)
+            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)
+            instructions = get_instructions(content_subject, content_grade, key_moments_text, socratic_mode)
+            print(f"=== instructions:{instructions} ===")
+            user_message_note = "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，絕對不要重複 user 的問句，請用引導的方式指引方向，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，請在回答的最後標註【參考：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內，回答時如果講到數學專有名詞，請用數學符號代替文字（Latex 用 $ 字號 render, ex: $x^2$)"
+            user_content = user_message + user_message_note
+            response_text, thread_id = handle_conversation_by_open_ai_assistant(client, user_content, instructions, assistant_id, thread_id, metadata, fallback=True)
 
     # 更新聊天历史
     chat_history = update_chat_history(user_message, response_text, chat_history)
@@ -2632,77 +2671,79 @@ def chat_with_opan_ai_assistant_streaming(user_message, chat_history, password,
     if chat_history is not None and len(chat_history) > CHAT_LIMIT:
         error_msg = f"此次對話超過上限（對話一輪{CHAT_LIMIT}次）"
         raise gr.Error(error_msg)
+    
+    print("===chat_with_opan_ai_assistant_streaming===")
+    print(user_message)
+    
 
-    try:
-        assistant_id = OPEN_AI_ASSISTANT_ID_GPT4 #GPT 4 turbo
-        # assistant_id = OPEN_AI_ASSISTANT_ID_GPT3 #GPT 3.5 turbo
-
-        client = OPEN_AI_CLIENT
-        # 直接安排逐字稿資料 in instructions
-        # if isinstance(trascript, str):
-        #     trascript_json = json.loads(trascript)
-        # else:
-        #     trascript_json = trascript
-        # trascript_text = json.dumps(trascript_json, ensure_ascii=False)
-        # # trascript_text 移除 \n, 空白
-        # trascript_text = trascript_text.replace("\n", "").replace(" ", "")
-
-        if isinstance(key_moments, str):
-            key_moments_json = json.loads(key_moments)
-        else:
-            key_moments_json = key_moments
-        # key_moments_json remove images
-        for moment in key_moments_json:
-            moment.pop('images', None)
-            moment.pop('end', None)
-            moment.pop('transcript', None)
-        key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)
-
-        instructions = get_instructions(content_subject, content_grade, key_moments_text, socratic_mode)
-        # 创建线程
-        if not thread_id:
-            thread = client.beta.threads.create()
-            thread_id = thread.id
-            print(f"new thread_id: {thread_id}")
-        else:
-            thread = client.beta.threads.retrieve(thread_id)
-            print(f"old thread_id: {thread_id}")
-        metadata = {
-            "youtube_id": video_id,
-            "user_data": user_data,
-            "content_subject": content_subject,
-            "content_grade": content_grade,
-            "assistant_id": assistant_id,
-            "is_streaming": "true",
-        }
-        client.beta.threads.update(
-            thread_id=thread_id,
-            metadata=metadata
-        )    
-        
-        # 向线程添加用户的消息
-        client.beta.threads.messages.create(
-            thread_id=thread.id,
-            role="user",
-            content=user_message + "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，不用提到「逐字稿」這個詞，用「內容」代替))，請在回答的最後標註【參考資料：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內）"
-        )
+    is_flagged, response_dict = get_chat_moderation(user_message)
+    assistant_id = OPEN_AI_ASSISTANT_ID_GPT4 #GPT 4 turbo
+    # assistant_id = OPEN_AI_ASSISTANT_ID_GPT3 #GPT 3.5 turbo
+    client = OPEN_AI_CLIENT
+    metadata = {
+        "youtube_id": video_id,
+        "user_data": user_data,
+        "content_subject": content_subject,
+        "content_grade": content_grade,
+        "assistant_id": assistant_id,
+        "is_streaming": "true",
+        "moderation_is_flagged": str(is_flagged),
+        # "moderation_response_dict": str(response_dict)
+    }
 
-        with client.beta.threads.runs.stream(
-            thread_id=thread.id,
-            assistant_id=assistant_id,
-            instructions=instructions,
-        ) as stream:
-            partial_messages = ""
-            for event in stream:
-                if event.data and event.data.object == "thread.message.delta":
-                    message = event.data.delta.content[0].text.value
-                    partial_messages += message
-                    yield partial_messages
+    if is_flagged:
+        partial_messages = "您的留言已被標記為不當內容，請重新發送。"
+        yield partial_messages
+    else:
+        try:
+            if isinstance(key_moments, str):
+                key_moments_json = json.loads(key_moments)
+            else:
+                key_moments_json = key_moments
+            # key_moments_json remove images
+            for moment in key_moments_json:
+                moment.pop('images', None)
+                moment.pop('end', None)
+                moment.pop('transcript', None)
+            key_moments_text = json.dumps(key_moments_json, ensure_ascii=False)
+
+            instructions = get_instructions(content_subject, content_grade, key_moments_text, socratic_mode)
+            # 创建线程
+            if not thread_id:
+                thread = client.beta.threads.create()
+                thread_id = thread.id
+                print(f"new thread_id: {thread_id}")
+            else:
+                thread = client.beta.threads.retrieve(thread_id)
+                print(f"old thread_id: {thread_id}")
+            
+            client.beta.threads.update(
+                thread_id=thread_id,
+                metadata=metadata
+            )    
+            
+            # 向线程添加用户的消息
+            client.beta.threads.messages.create(
+                thread_id=thread.id,
+                role="user",
+                content=user_message + "/n 請嚴格遵循instructions，擔任一位蘇格拉底家教，請一定要用繁體中文回答 zh-TW，並用台灣人的禮貌口語表達，回答時不要特別說明這是台灣人的語氣，不用提到「逐字稿」這個詞，用「內容」代替))，請在回答的最後標註【參考資料：（時）:（分）:（秒）】，（如果是反問學生，就只問一個問題，請幫助學生更好的理解資料，字數在100字以內）"
+            )
 
-    except Exception as e:
-        print(f"Error: {e}")
-        raise gr.Error(f"Error: {e}")
-    
+            with client.beta.threads.runs.stream(
+                thread_id=thread.id,
+                assistant_id=assistant_id,
+                instructions=instructions,
+            ) as stream:
+                partial_messages = ""
+                for event in stream:
+                    if event.data and event.data.object == "thread.message.delta":
+                        message = event.data.delta.content[0].text.value
+                        partial_messages += message
+                        yield partial_messages
+        except Exception as e:
+            print(f"Error: {e}")
+            raise gr.Error(f"Error: {e}")
+        
     user_id = user_data
     route = "chat_with_opan_ai_assistant_streaming"
     endpoint = "assistant_streaming"

local_config_example.json

@@ -7,6 +7,7 @@
   "OPEN_AI_KEY": "sk-proj-xxxxxxxxxx",
   "OPEN_AI_ASSISTANT_ID_GPT4_BOT1": "asst_3cxxxxxxxxxxjrQio9",
   "OPEN_AI_ASSISTANT_ID_GPT3_BOT1": "asst_mcuxxxxxxxxxx5L4e",
+  "OPEN_AI_MODERATION_BOT1": "sk-proj-xxxxxxxxxx",
   "OPEN_AI_KEY_BOT2": "sk-proj-5HsYxxxxxxxxxxkH4pShu",
   "OPEN_AI_ASSISTANT_ID_GPT4_BOT2": "asst_3xxxxxxxxxxrQio9",
   "OPEN_AI_ASSISTANT_ID_GPT3_BOT2": "asst_mxxxxxxxxxx6vg5L4e",