generate_topic_sentence_feedback

app.py

@@ -286,89 +286,148 @@ def update_points_input(points):
 
 def generate_topic_sentences(model, max_tokens, sys_content, scenario, eng_level, topic, points, user_generate_topic_sentences_prompt):
     """
-    根据系统提示和用户输入的情境及要点，调用OpenAI API生成相关的主题句及其合理性解释。
+    根据系统提示和用户输入的情境及要点，调用 LLM API 生成相关的主题句及其合理性解释。
     """
+    try:
+        exam_history_prompt = ""
+        if eng_level == "台灣學科能力測驗等級":
+            exam_history = get_exam_history()
+            exam_history_prompt = f"""
+                Please refer a topic scenario from the following exam history:
+                {exam_history}
+                give similar topic scenario and level of English. But don't use the same topic scenario.
+            """
 
-    if eng_level == "台灣學科能力測驗等級":
-        exam_history = get_exam_history()
-        exam_history_prompt = f"""
-            Please refer a topic scenario from the following exam history:
-            {exam_history}
-            give similar topic scenario and level of English. But don't use the same topic scenario.
+        user_content = f"""
+            scenario is: {scenario}
+            english level is: {eng_level}
+            topic is: {topic}
+            points is: {points}
+            ---
+            exam_history_prompt: {exam_history_prompt}
+            ---
+            {user_generate_topic_sentences_prompt}
         """
-    else:
-        exam_history_prompt = ""
+        
+        messages = [
+            {"role": "system", "content": sys_content},
+            {"role": "user", "content": user_content}
+        ]
 
-    user_content = f"""
-        scenario is: {scenario}
-        english level is: {eng_level}
-        topic is: {topic}
-        points is: {points}
-        ---
-        exam_history_prompt: {exam_history_prompt}
-        ---
-        {user_generate_topic_sentences_prompt}
-    """
-    messages = [
-        {"role": "system", "content": sys_content},
-        {"role": "user", "content": user_content}
-    ]
-    response_format = { "type": "json_object" }
+        # 根據模型選擇 provider
+        if "gemini" in model.lower():
+            print("====gemini====")
+            provider = GeminiProvider()
+        else:
+            print("====openai====")
+            provider = OpenAIProvider(OPEN_AI_CLIENT)
 
-    request_payload = {
-        "model": model,
-        "messages": messages,
-        "max_tokens": max_tokens,
-        "response_format": response_format
-    }
+        # 使用 LLMService 處理請求
+        llm_service = LLMService(provider)
+        content = llm_service.chat(
+            prompt=f"{sys_content}\n{user_content}" if "gemini" in model.lower() else None,
+            messages=messages,
+            model=model,
+            max_tokens=max_tokens,
+            response_format={"type": "json_object"}
+        )
+
+        print(f"====generate_topic_sentences====")
+        print(content)
+
+        # 處理回應格式
+        try:
+            # 如果回應包含多個 JSON 區塊，嘗試找出正確的那個
+            if isinstance(content, str):
+                # 移除可能的 markdown 格式
+                if "```json" in content:
+                    json_blocks = content.split("```json")
+                    for block in json_blocks:
+                        if "```" in block:
+                            content = block.split("```")[0].strip()
+                        else:
+                            content = block.strip()
+                
+                # 清理內容，確保是有效的 JSON
+                content = content.strip()
+                if content.startswith("```") and content.endswith("```"):
+                    content = content[3:-3].strip()
+                
+                print(f"Cleaned content: {content}")
+                response_content = json.loads(content)
+            else:
+                response_content = content
+
+            json_content = response_content["results"]
+            topic_sentences_list = [item["topic-sentence"] for item in json_content]
+            random.shuffle(topic_sentences_list)
+            
+            gr_update_json = gr.update(value=json_content)
+            gr_update_radio = gr.update(choices=topic_sentences_list, visible=True)
+            return gr_update_json, gr_update_radio
+            
+        except (json.JSONDecodeError, KeyError, ValueError) as e:
+            print(f"Error parsing topic sentences: {e}")
+            print(f"Content causing error: {content}")
+            raise gr.Error("無法解析主題句，請重新嘗試")
 
-    try:
-        response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
-        response_content = json.loads(response.choices[0].message.content)
-        json_content = response_content["results"]
-        topic_sentences_list = [item["topic-sentence"] for item in json_content]
-        random.shuffle(topic_sentences_list)
-        
-        gr_update_json = gr.update(value=json_content)
-        gr_update_radio = gr.update(choices=topic_sentences_list, visible=True)
     except Exception as e:
         print(f"An error occurred while generating topic sentences: {e}")
-        raise gr.Error("網路塞車，請重新嘗試一次！")
-
-    return gr_update_json, gr_update_radio
+        error_msg = "網路塞車，請重新嘗試一次！"
+        if "rate limit" in str(e).lower():
+            error_msg = "請求過於頻繁，請稍後再試"
+        elif "invalid_request_error" in str(e).lower():
+            error_msg = "請求格式錯誤，請檢查輸入"
+        raise gr.Error(error_msg)
 
 def generate_topic_sentence_feedback(model, max_tokens, sys_content, scenario, eng_level, topic, points, topic_sentence, user_generate_topic_sentence_feedback_prompt):
     """
-    根据系统提示和用户输入的情境、主题、要点、主题句，调用OpenAI API生成相关的主题句反饋。
-    """
-    user_content = f"""
-        scenario is: {scenario}
-        english level is: {eng_level}
-        topic is: {topic}
-        points is: {points}
-        ---
-        my written topic sentence is: {topic_sentence}
-        ---
-        {user_generate_topic_sentence_feedback_prompt}
+    根据系统提示和用户输入的情境、主题、要点、主题句，调用 LLM API 生成相关的主题句反饋。
     """
-    messages = [
-        {"role": "system", "content": sys_content},
-        {"role": "user", "content": user_content}
-    ]
+    try:
+        user_content = f"""
+            scenario is: {scenario}
+            english level is: {eng_level}
+            topic is: {topic}
+            points is: {points}
+            ---
+            my written topic sentence is: {topic_sentence}
+            ---
+            {user_generate_topic_sentence_feedback_prompt}
+        """
+        
+        messages = [
+            {"role": "system", "content": sys_content},
+            {"role": "user", "content": user_content}
+        ]
 
-    request_payload = {
-        "model": model,
-        "messages": messages,
-        "max_tokens": max_tokens,
-    }
+        # 根據模型選擇 provider
+        if "gemini" in model.lower():
+            print("====gemini====")
+            provider = GeminiProvider()
+        else:
+            print("====openai====")
+            provider = OpenAIProvider(OPEN_AI_CLIENT)
+
+        # 使用 LLMService 處理請求
+        llm_service = LLMService(provider)
+        content = llm_service.chat(
+            prompt=f"{sys_content}\n{user_content}" if "gemini" in model.lower() else None,
+            messages=messages,
+            model=model,
+            max_tokens=max_tokens
+        )
 
-    try:
-        response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
-        content = response.choices[0].message.content.strip()
         gr_update = gr.update(value=content, visible=True)
+
     except Exception as e:
-        print(f"An error occurred while generating topic sentence feedback: {e}")
-        raise gr.Error("網路塞車，請重新嘗試一次！")
+        print(f"An error occurred while generating topic sentence feedback: {str(e)}")
+        error_msg = "網路塞車，請重新嘗試一次！"
+        if "rate limit" in str(e).lower():
+            error_msg = "請求過於頻繁，請稍後再試"
+        elif "invalid_request_error" in str(e).lower():
+            error_msg = "請求格式錯誤，請檢查輸入"
+        raise gr.Error(error_msg)
 
     return gr_update