generate_points

app.py

@@ -1,4 +1,3 @@
-
 import gradio as gr
 from openai import OpenAI
 
@@ -26,6 +25,7 @@ from assignment_ui import create_assignment_ui
 from assignment_service import AssignmentService
 from submission_service import SubmissionService
 from dashboard_service import DashboardService
+from llms import GeminiProvider, OpenAIProvider, LLMService
 
 is_env_local = os.getenv("IS_ENV_LOCAL", "false") == "true"
 print(f"is_env_local: {is_env_local}")
@@ -72,7 +72,6 @@ vertexai.init(
     credentials=google_creds,
 )
 
-
 _AssignmentService = AssignmentService(GCS_SERVICE)
 _SubmissionService = SubmissionService(GCS_SERVICE)
 _DashboardService = DashboardService(_AssignmentService, _SubmissionService)
@@ -158,98 +157,129 @@ def get_exam_history():
 
 def generate_topics(model, max_tokens, sys_content, scenario, eng_level, user_generate_topics_prompt):
     """
-    根据系统提示和用户输入的情境及主题，调用OpenAI API生成相关的主题句。
-    """
-
-    exam_history = get_exam_history()
-    exam_history_prompt = f"""
-        Please refer a topic scenario from the following exam history: 
-        {exam_history}
-        Base on English level to give similar topic scenario. But don't use the same topic scenario.
-    """
-
-    user_content = f"""
-        english level is: {eng_level}
-        ---
-        exam_history_prompt: {exam_history_prompt}
-        ---
-        {user_generate_topics_prompt}
+    根据系统提示和用户输入的情境及主题，调用 LLM API 生成相关的主题句。
     """
-    messages = [
-        {"role": "system", "content": sys_content},
-        {"role": "user", "content": user_content}
-    ]
+    try:
+        exam_history = get_exam_history()
+        exam_history_prompt = f"""
+            Please refer a topic scenario from the following exam history: 
+            {exam_history}
+            Base on English level to give similar topic scenario. But don't use the same topic scenario.
+        """
 
-    request_payload = {
-        "model": model,
-        "messages": messages,
-        "max_tokens": max_tokens,
-        "response_format": { "type": "json_object" }
-    }
+        user_content = f"""
+            english level is: {eng_level}
+            ---
+            exam_history_prompt: {exam_history_prompt}
+            ---
+            {user_generate_topics_prompt}
+        """
+        
+        messages = [
+            {"role": "system", "content": sys_content},
+            {"role": "user", "content": user_content}
+        ]
 
-    try:
-        # response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
-        # content = response.choices[0].message.content
-        # topics = json.loads(content)["topics"]
-        # print(f"====generate_topics====")
-        # print(topics)
-        # gr_update = gr.update(choices=topics, visible=True)
-    
-        model_name = "gemini-1.5-pro"
-        gemini_model = GenerativeModel(model_name=model_name)
-        model_response = gemini_model.generate_content(
-            f"{sys_content}, {user_content}"
+        # 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,
+            response_format={"type": "json_object"}
         )
-        content = model_response.candidates[0].content.parts[0].text
-        print(f"====generate_topics====")
-        print(content)
-        print("=====")
-        if "```json" in content:
+
+        # 處理回應格式
+        if isinstance(content, str) and "```json" in content:
             content = content.replace("```json", "").replace("```", "")
-        topics = json.loads(content)["topics"]
-        gr_update = gr.update(choices=topics, visible=True)
-    except Exception as e:
-        print(f"An error occurred while generating topics: {e}")
-        raise gr.Error("網路塞車，請重新嘗試一次！")
+            
+        try:
+            topics = json.loads(content)["topics"]
+            if not topics or not isinstance(topics, list):
+                raise ValueError("Invalid topics format")
+            return gr.update(choices=topics, visible=True)
+        except (json.JSONDecodeError, KeyError, ValueError) as e:
+            print(f"Error parsing topics: {e}")
+            raise gr.Error("無法解析主題，請重新嘗試")
 
-    return gr_update
+    except Exception as e:
+        print(f"An error occurred while generating topics: {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)
 
 def update_topic_input(topic):
     return topic
 
 def generate_points(model, max_tokens, sys_content, scenario, eng_level, topic, user_generate_points_prompt):
     """
-    根据系统提示和用户输入的情境、主题，调用OpenAI API生成相关的主题句。
-    """
-    user_content = f"""
-        scenario is: {scenario}
-        english level is: {eng_level}
-        topic is: {topic}
-        ---
-        {user_generate_points_prompt}
+    根据系统提示和用户输入的情境、主题，调用 LLM API 生成相关的主题句。
     """
-    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}
+            ---
+            {user_generate_points_prompt}
+        """
+        
+        messages = [
+            {"role": "system", "content": sys_content},
+            {"role": "user", "content": user_content}
+        ]
 
-    request_payload = {
-        "model": model,
-        "messages": messages,
-        "max_tokens": max_tokens,
-        "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)
+
+        # 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"}
+        )
 
-    try:
-        response = OPEN_AI_CLIENT.chat.completions.create(**request_payload)
-        content = response.choices[0].message.content
-        points = json.loads(content)["points"]
-        gr_update = gr.update(choices=points, visible=True)
-    except Exception as e:
-        print(f"An error occurred while generating points: {e}")
-        raise gr.Error("網路塞車，請重新嘗試一次！")
+        # 處理回應格式
+        if isinstance(content, str) and "```json" in content:
+            content = content.replace("```json", "").replace("```", "")
+            
+        try:
+            points = json.loads(content)["points"]
+            if not points or not isinstance(points, list):
+                raise ValueError("Invalid points format")
+            return gr.update(choices=points, visible=True)
+        except (json.JSONDecodeError, KeyError, ValueError) as e:
+            print(f"Error parsing points: {e}")
+            raise gr.Error("無法解析要點，請重新嘗試")
 
-    return gr_update
+    except Exception as e:
+        print(f"An error occurred while generating points: {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)
 
 def update_points_input(points):
     return points
@@ -1401,7 +1431,7 @@ def get_chinese_paragraph_1st_evaluate_content(
 
         # Output format:
         1. 先給 綜合回饋、評分標準與回饋、修改範例
-        2. 再將評分標準與回饋的內容以JSON格式輸出，並且請使用繁體中文（ZH-TW）來評分段落並輸出，用 ```json ..... ``` 包裹:
+        2. 再將評分標準與回饋的容以JSON格式輸出，並且請使用繁體中文（ZH-TW）來評分段落並輸出，用 ```json ..... ``` 包裹:
         3. please use Chinese language (ZH-TW) to evaluate the paragraph and output use JSON format
         4. json 內容只需要給「主題與內容」、「遣詞造句」的評分標準與回饋
 
@@ -2098,8 +2128,8 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                                 gr.Image(value="https://storage.googleapis.com/jutor/jutor_en_chinese.jpg", show_label=False, show_download_button=False)
 
                     # ===== 基礎級使用者 =====
-                    with gr.Row(visible=False) as default_params:
-                        model = gr.Radio(["gpt-4o", "gpt-4-turbo"], label="Model", value="gpt-4o")
+                    with gr.Row(visible=True) as default_params:
+                        model = gr.Radio(["gpt-4o", "gpt-4-turbo", "gemini-1.5-pro"], label="Model", value="gemini-1.5-pro")
                         max_tokens = gr.Slider(minimum=50, maximum=4000, value=4000, label="Max Tokens")
                         sys_content_input = gr.Textbox(label="System Prompt", value="You are an English teacher who is practicing with me to improve my English writing skill.")
                     with gr.Row():
@@ -2168,6 +2198,10 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                                     Give me 10 randon topics, 
                                     for a paragraph. Just the topics, no explanation, use English language base on  eng_level. 
                                     Make sure the vocabulary you use is at eng_level.
+                                    more funny and interesting topics are better. 
+                                    such as my favorite cartoon/anime, my favorite food, my favorite movie, etc.
+                                    but not always about yourself. Maybe like "Games at school" or "Being a vegetarian".
+                                    You can give some topics start with My ..., but not every topic use the same prefix.
                                     output use JSON 
                                     EXAMPLE:
                                     "topics":["topic1", "topic2", "topic3", "topic4", "topic5", "topic6", "topic7", "topic8", "topic9", "topic10"]
@@ -2215,10 +2249,11 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                                     Please provide main points to develop in a paragraph about topic in the context of scenario, 
                                     use simple English language and make sure the vocabulary you use is at eng_level.
                                     No more explanation either no developing these points into a simple paragraph.
+                                    At least 5 points/keywords and complete the paragraph.
                                     Output use JSON format
                                     EXAMPLE:
                                     {{
-                                        "points":["point1", "point2", "point3"]
+                                        "points":["point1", "point2", "point3", "point4", "point5"]
                                     }}
                                 """  
                                 user_generate_points_prompt = gr.Textbox(label="Points Prompt", value=default_generate_points_prompt, visible=False)
@@ -2690,11 +2725,11 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
 
                                     Final Output JSON Format:
                                     {{
-                                    “content“: {{content’s dict}},
-                                    “organization“: {{organization'dict}},
-                                    “grammar_and_usage“: {{grammar_and_usage'dict}},
-                                    “vocabulary“: {{vocabulary'dict}},
-                                    “coherence_and_cohesion“: {{coherence_and_cohesion'dict}}
+                                    "content": {{content's dict}},
+                                    "organization": {{organization'dict}},
+                                    "grammar_and_usage": {{grammar_and_usage'dict}},
+                                    "vocabulary": {{vocabulary'dict}},
+                                    "coherence_and_cohesion": {{coherence_and_cohesion'dict}}
                                     }}            
                                 """
                                 user_generate_paragraph_evaluate_prompt = gr.Textbox(label="Paragraph evaluate Prompt", value=default_user_generate_paragraph_evaluate_prompt, visible=False)
@@ -3029,7 +3064,7 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                         with gr.Column():
                             generate_full_paragraph_refine_button = gr.Button("✨ JUTOR 段落改善建議", variant="primary")
                             full_paragraph_refine_output_table = gr.DataFrame(label="段落改善建議", wrap=True, interactive=False)
-                            full_paragraph_refine_output = gr.HTML(label="修改建議", visible=False)
+                            full_paragraph_refine_output = gr.HTML(label="改建議", visible=False)
                             gr.Markdown("## 修改結果")
                             full_paragraph_refine_output_diff = gr.HTML()
 
@@ -3574,7 +3609,7 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                         chinese_full_paragraph_refine_input = gr.TextArea(label="這是你的原始寫作內容，參考建議，你可以選擇是否修改：", show_copy_button=True)
                     with gr.Column():
                         with gr.Row():
-                            generate_chinese_full_paragraph_refine_button = gr.Button("段落全文分析", variant="primary")
+                            generate_chinese_full_paragraph_refine_button = gr.Button("���落全文分析", variant="primary")
                         with gr.Row():
                             chinese_full_paragraph_refine_output_text = gr.Markdown(label="段落全文分析")
                         with gr.Row():
@@ -3741,7 +3776,7 @@ with gr.Blocks(theme=THEME, css=CSS) as demo:
                                     gr.Markdown("<span style='color:#4e80ee'>作業繳交截止時間</span>")
                                     chinese_assignment_submission_deadline_history_log = gr.Markdown()
                                     gr.Markdown("---")
-                                    gr.Markdown("# 回傳作業內容")
+                                    gr.Markdown("# 回傳作業容")
 
                             with gr.Row():
                                 with gr.Column():