update

app.py

@@ -489,6 +489,7 @@ def process_transcript_and_screenshots_on_gcs(video_id):
     transcript_file_name = f'{video_id}_transcript.json'
     transcript_blob_name = f"{video_id}/{transcript_file_name}"
     # 检查逐字稿是否存在
+    is_new_transcript = False
     is_transcript_exists = GCS_SERVICE.check_file_exists(bucket_name, transcript_blob_name)
     if not is_transcript_exists:
         # 从YouTube获取逐字稿并上传
@@ -507,6 +508,7 @@ def process_transcript_and_screenshots_on_gcs(video_id):
 
         transcript_text = json.dumps(transcript, ensure_ascii=False, indent=2)
         upload_file_to_gcs_with_json_string(gcs_client, bucket_name, transcript_blob_name, transcript_text)
+        is_new_transcript = True
     else:
         # 逐字稿已存在，下载逐字稿内容
         print("逐字稿已存在于GCS中")
@@ -542,15 +544,20 @@ def process_transcript_and_screenshots_on_gcs(video_id):
             img_file_id = upload_img_and_get_public_url(gcs_client, bucket_name, screenshot_blob_name, screenshot_path)
             entry['img_file_id'] = img_file_id
             print(f"截图已上传到GCS: {img_file_id}")
-        
-    # 更新逐字稿文件
-    print("===更新逐字稿文件===")
-    print(transcript)
-    print("===更新逐字稿文件===")
-    updated_transcript_text = json.dumps(transcript, ensure_ascii=False, indent=2)
-    upload_file_to_gcs_with_json_string(gcs_client, bucket_name, transcript_blob_name, updated_transcript_text)
-    print("逐字稿已更新，包括截图链接")
-    updated_transcript_json = json.loads(updated_transcript_text)
+            is_new_transcript = True
+    
+    # 確認是否更新逐字稿文件
+    if is_new_transcript:
+        # 更新逐字稿文件
+        print("===更新逐字稿文件===")
+        print(transcript)
+        print("===更新逐字稿文件===")
+        updated_transcript_text = json.dumps(transcript, ensure_ascii=False, indent=2)
+        upload_file_to_gcs_with_json_string(gcs_client, bucket_name, transcript_blob_name, updated_transcript_text)
+        print("逐字稿已更新，包括截图链接")
+        updated_transcript_json = json.loads(updated_transcript_text)
+    else:
+        updated_transcript_json = transcript
 
     return updated_transcript_json
 
@@ -691,11 +698,11 @@ def screenshot_youtube_video(youtube_id, snapshot_sec):
 
 
 # ---- Web ----
-def process_web_link(link):
-    # 抓取和解析网页内容
-    response = requests.get(link)
-    soup = BeautifulSoup(response.content, 'html.parser')
-    return soup.get_text()
+# def process_web_link(link):
+#     # 抓取和解析网页内容
+#     response = requests.get(link)
+#     soup = BeautifulSoup(response.content, 'html.parser')
+#     return soup.get_text()
 
 
 # ---- LLM Generator ----
@@ -1766,6 +1773,104 @@ def poll_run_status(run_id, thread_id, timeout=600, poll_interval=5):
     
     return run.status
 
+def streaming_chat_with_open_ai(user_message, chat_history, password, thread_id, trascript, content_subject, content_grade):
+    verify_password(password)
+
+    print("===streaming_chat_with_open_ai===")
+    print(thread_id)
+    
+    # 先計算 user_message 是否超過 500 個字
+    if len(user_message) > 1500:
+        error_msg = "你的訊息太長了，請縮短訊息長度至五百字以內"
+        raise gr.Error(error_msg)
+    
+    # 如果 chat_history 超過 10 則訊息，直接 return "對話超過上限"
+    if chat_history is not None and len(chat_history) > 10:
+        error_msg = "此次對話超過上限"
+        raise gr.Error(error_msg)
+
+    # fake data
+    socratic_mode = True
+
+    try:
+        assistant_id = "asst_kmvZLNkDUYaNkMNtZEAYxyPq"
+        client = OPEN_AI_CLIENT
+        # 直接安排逐字稿資料 in instructions
+        trascript_json = json.loads(trascript)
+        # 移除 embed_url, screenshot_path
+        for entry in trascript_json:
+            entry.pop('embed_url', None)
+            entry.pop('screenshot_path', None)
+        trascript_text = json.dumps(trascript_json, ensure_ascii=False, indent=2)
+        # trascript_text 移除 \n, 空白
+        trascript_text = trascript_text.replace("\n", "").replace(" ", "")
+
+        instructions = f"""
+            科目：{content_subject}
+            年級：{content_grade}
+            逐字稿資料：{trascript_text}
+            -------------------------------------
+            你是一個專業的{content_subject}老師， user 為{content_grade}的學生
+            socratic_mode = {socratic_mode}
+            if socratic_mode is True，
+            - 請用蘇格拉底式的提問方式，引導學生思考，並且給予學生一些提示
+            - 一次只問一個問題，字數在100字以內
+            - 不要直接給予答案，讓學生自己思考
+            - 但可以給予一些提示跟引導，例如給予影片的時間軸，讓學生自己去找答案
+            
+            if socratic_mode is False，
+            - 直接回答學生問題，字數在100字以內
+
+            rule:
+            - 請一定要用繁體中文回答 zh-TW，並用台灣人的口語表達，回答時不用特別說明這是台灣人的語氣，也不用說這是「台語的說法」
+            - 不用提到「逐字稿」這個詞，用「內容」代替
+            - 如果學生問了一些問題你無法判斷，請告訴學生你無法判斷，並建議學生可以問其他問題
+            - 或者你可以反問學生一些問題，幫助學生更好的理解資料，字數在100字以內
+            - 如果學生的問題與資料文本無關，請告訴學生你「無法回答超出影片範圍的問題」，並告訴他可以怎麼問什麼樣的問題（一個就好）
+            - 只要是參考逐字稿資料，請在回答的最後標註【參考資料：（分）：（秒）】
+            - 回答範圍一定要在逐字稿資料內，不要引用其他資料，請嚴格執行
+            - 並在重複問句後給予學生鼓勵，讓學生有學習的動力
+            - 請用 {content_grade} 的學生能懂的方式回答
+        """
+
+        # 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}")
+        
+        # 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）"
+        )
+
+        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}")
+
+def create_thread_id():
+    thread = OPEN_AI_CLIENT.beta.threads.create()
+    thread_id = thread.id
+    print(f"create new thread_id: {thread_id}")
+    return thread_id
+
 # --- Slide mode ---
 def update_slide(direction):
     global TRANSCRIPTS
@@ -1794,6 +1899,8 @@ def prev_slide():
 def next_slide():
     return update_slide(1)
 
+
+# --- Init params ---
 def init_params(text, request: gr.Request):
     if request:
         print("Request headers dictionary:", request.headers)
@@ -1826,6 +1933,21 @@ def init_params(text, request: gr.Request):
 
     return admin, reading_passage_admin, summary_admin, see_detail, password_text, youtube_link
 
+def update_state(content_subject, content_grade, trascript, question_1, question_2, question_3):
+    # inputs=[content_subject, content_grade, df_string_output],
+    # outputs=[content_subject_state, content_grade_state, trascript_state]
+    content_subject_state = content_subject
+    content_grade_state = content_grade
+    trascript_state = trascript
+    streaming_chat_thread_id_state = create_thread_id()
+    ai_chatbot_question_1 = question_1
+    ai_chatbot_question_2 = question_2
+    ai_chatbot_question_3 = question_3
+
+    return content_subject_state, content_grade_state, trascript_state, streaming_chat_thread_id_state, \
+        ai_chatbot_question_1, ai_chatbot_question_2, ai_chatbot_question_3
+
+
 HEAD = """
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
@@ -1893,15 +2015,20 @@ HEAD = """
 with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, secondary_hue=gr.themes.colors.amber, text_size = gr.themes.sizes.text_lg), head=HEAD) as demo:
     with gr.Row() as admin:
         password = gr.Textbox(label="Password", type="password", elem_id="password_input", visible=True)
-        file_upload = gr.File(label="Upload your CSV or Word file", visible=False)
         youtube_link = gr.Textbox(label="Enter YouTube Link", elem_id="youtube_link_input", visible=True)
         video_id = gr.Textbox(label="video_id", visible=False)
-        web_link = gr.Textbox(label="Enter Web Page Link", visible=False)
+        # file_upload = gr.File(label="Upload your CSV or Word file", visible=False)
+        # web_link = gr.Textbox(label="Enter Web Page Link", visible=False)
         user_data = gr.Textbox(label="User Data", elem_id="user_data_input", visible=True)
         youtube_link_btn = gr.Button("Submit_YouTube_Link", elem_id="youtube_link_btn", visible=True)
+    with gr.Row() as data_state:
+        content_subject_state = gr.State()  # 使用 gr.State 存储 content_subject
+        content_grade_state = gr.State()  # 使用 gr.State 存储 content_grade
+        trascript_state = gr.State()  # 使用 gr.State 存储 trascript
+        streaming_chat_thread_id_state = gr.State()  # 使用 gr.State 存储 streaming_chat_thread_id
     with gr.Tab("AI小精靈"):
         with gr.Row():
-            with gr.Tab("飛特"):
+            with gr.Tab("飛特精靈"):
                 bot_avatar = "https://junyi-avatar.s3.ap-northeast-1.amazonaws.com/live/%20%20foxcat-star-18.png?v=20231113095823614"
                 user_avatar = "https://junyitopicimg.s3.amazonaws.com/s4byy--icon.jpe?v=20200513013523726"
                 latex_delimiters = [{"left": "$", "right": "$", "display": False}]
@@ -1917,28 +2044,33 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
                         btn_create_question = gr.Button("生成其他問題", variant="primary")
                     openai_chatbot_audio_input = gr.Audio(sources=["microphone"], type="filepath")
                 with gr.Row():
-                        msg = gr.Textbox(label="訊息",scale=3)
-                        send_button = gr.Button("送出", variant="primary", scale=1)
-            # with gr.Tab("GROQ"):
-            #     groq_ai_name = gr.Textbox(label="AI 助理名稱", value="groq", visible=False)
-            #     groq_chatbot = gr.Chatbot(avatar_images=[bot_avatar, user_avatar], label="groq mode chatbot", show_share_button=False, likeable=True)
-            #     groq_msg = gr.Textbox(label="Message")
-            #     groq_send_button = gr.Button("Send", variant="primary")
-            # with gr.Tab("JUTOR"):
-            #     jutor_ai_name = gr.Textbox(label="AI 助理名稱", value="jutor", visible=False)
-            #     jutor_chatbot = gr.Chatbot(avatar_images=[bot_avatar, user_avatar], label="jutor mode chatbot", show_share_button=False, likeable=True)
-            #     jutor_msg = gr.Textbox(label="Message")
-            #     jutor_send_button = gr.Button("Send", variant="primary")
-            # with gr.Tab("CLAUDE"):
-            #     claude_ai_name = gr.Textbox(label="AI 助理名稱", value="claude3", visible=False)
-            #     claude_chatbot = gr.Chatbot(avatar_images=[bot_avatar, user_avatar], label="claude mode chatbot", show_share_button=False, likeable=True)
-            #     claude_msg = gr.Textbox(label="Message")
-            #     claude_send_button = gr.Button("Send", variant="primary")
+                    msg = gr.Textbox(label="訊息",scale=3)
+                    send_button = gr.Button("送出", variant="primary", scale=1)
+            with gr.Tab("飛特音速"):
+                additional_inputs = [password, streaming_chat_thread_id_state, trascript_state, content_subject_state, content_grade_state]
+                streaming_chat = gr.ChatInterface(
+                    fn=streaming_chat_with_open_ai,
+                    additional_inputs=additional_inputs,
+                    submit_btn="送出",
+                    retry_btn=None,
+                    undo_btn="⏪ 上一步",
+                    clear_btn="🗑️ 清除全部",
+                    stop_btn="🛑 停止",
+                )
             with gr.Tab("其他精靈"):
                 ai_name = gr.Dropdown(label="選擇 AI 助理", choices=["jutor", "claude3", "groq"], value="jutor")
-                ai_chatbot = gr.Chatbot(avatar_images=[bot_avatar, user_avatar], label="ai_chatbot", show_share_button=False, likeable=True, show_label=False)
-                ai_msg = gr.Textbox(label="Message")
-                ai_send_button = gr.Button("Send", variant="primary")    
+                ai_chatbot = gr.Chatbot(avatar_images=[bot_avatar, user_avatar], label="ai_chatbot", show_share_button=False, likeable=True, show_label=False, latex_delimiters=latex_delimiters)
+                ai_chatbot_socratic_mode_btn = gr.Checkbox(label="蘇格拉底家教助理模式", value=True, visible=False)
+                with gr.Row():
+                    with gr.Accordion("你也有類似的問題想問嗎？", open=False) as ask_questions_accordion_2:
+                        ai_chatbot_question_1 = gr.Button("問題一")
+                        ai_chatbot_question_2 = gr.Button("問題一")
+                        ai_chatbot_question_3 = gr.Button("問題一")
+                    ai_chatbot_audio_input = gr.Audio(sources=["microphone"], type="filepath")
+                with gr.Row():
+                    ai_msg = gr.Textbox(label="Message",scale=3)
+                    ai_send_button = gr.Button("Send", variant="primary",scale=1)    
+            
     with gr.Tab("文章模式"):
         with gr.Row() as reading_passage_admin:
             reading_passage_kind = gr.Textbox(value="reading_passage", show_label=False)
@@ -1962,6 +2094,7 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
     with gr.Tab("關鍵時刻"):
         with gr.Row():
             key_moments_html = gr.HTML(value="")
+
     with gr.Tab("教學備課"):
         with gr.Row():
             content_subject = gr.Dropdown(label="選擇主題", choices=["數學", "自然", "國文", "英文", "社會","物理", "化學", "生物", "地理", "歷史", "公民"], value="", visible=False)             
@@ -2070,7 +2203,7 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
             mind_map_html = gr.HTML()
        
     # --- Event ---
-    # OPENAI 模式
+    # OPENAI ASSISTANT CHATBOT 模式
     send_button.click(
         chat_with_opan_ai_assistant,
         inputs=[password, video_id, thread_id, df_string_output, msg, chatbot, content_subject, content_grade, socratic_mode_btn],
@@ -2081,35 +2214,10 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
         inputs=[password, openai_chatbot_audio_input],
         outputs=[msg]
     )
-    # # GROQ 模式
-    # groq_send_button.click(
-    #     chat_with_ai,
-    #     inputs=[groq_ai_name, password, video_id, df_string_output, groq_msg, groq_chatbot, content_subject, content_grade, socratic_mode_btn],
-    #     outputs=[groq_msg, groq_chatbot]
-    # )
-    # # JUTOR API 模式
-    # jutor_send_button.click(
-    #     chat_with_ai,
-    #     inputs=[jutor_ai_name, password, video_id, df_string_output, jutor_msg, jutor_chatbot, content_subject, content_grade, socratic_mode_btn],
-    #     outputs=[jutor_msg, jutor_chatbot]
-    # )
-    # # CLAUDE 模式
-    # claude_send_button.click(
-    #     chat_with_ai,
-    #     inputs=[claude_ai_name, password, video_id, df_string_output, claude_msg, claude_chatbot, content_subject, content_grade, socratic_mode_btn],
-    #     outputs=[claude_msg, claude_chatbot]
-    # )
-    # ai_chatbot 模式
-    ai_send_button.click(
-        chat_with_ai,
-        inputs=[ai_name, password, video_id, df_string_output, ai_msg, ai_chatbot, content_subject, content_grade, socratic_mode_btn],
-        outputs=[ai_msg, ai_chatbot]
-    )
-
-    # 连接按钮点击事件
-    btn_1_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_1, chatbot, content_subject, content_grade, socratic_mode_btn]
-    btn_2_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_2, chatbot, content_subject, content_grade, socratic_mode_btn]
-    btn_3_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_3, chatbot, content_subject, content_grade, socratic_mode_btn]
+    # OPENAI ASSISTANT CHATBOT 連接按鈕點擊事件
+    btn_1_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_1, chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
+    btn_2_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_2, chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
+    btn_3_chat_with_opan_ai_assistant_input =[password, video_id, thread_id, df_string_output, btn_3, chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
     btn_1.click(
         chat_with_opan_ai_assistant,
         inputs=btn_1_chat_with_opan_ai_assistant_input, 
@@ -2125,18 +2233,44 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
         inputs=btn_3_chat_with_opan_ai_assistant_input, 
         outputs=[msg, chatbot, thread_id]
     )
-
     btn_create_question.click(
         change_questions, 
         inputs = [password, df_string_output], 
         outputs = [btn_1, btn_2, btn_3]
     )
 
+    # ai_chatbot 模式
+    ai_send_button.click(
+        chat_with_ai,
+        inputs=[ai_name, password, video_id, df_string_output, ai_msg, ai_chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn],
+        outputs=[ai_msg, ai_chatbot]
+    )
+    # ai_chatbot 连接按钮点击事件
+    ai_chatbot_question_1_chat_with_ai_input =[ai_name, password, video_id, df_string_output, ai_chatbot_question_1, ai_chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
+    ai_chatbot_question_2_chat_with_ai_input =[ai_name, password, video_id, df_string_output, ai_chatbot_question_2, ai_chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
+    ai_chatbot_question_3_chat_with_ai_input =[ai_name, password, video_id, df_string_output, ai_chatbot_question_3, ai_chatbot, content_subject, content_grade, ai_chatbot_socratic_mode_btn]
+    ai_chatbot_question_1.click(
+        chat_with_ai,
+        inputs=ai_chatbot_question_1_chat_with_ai_input,
+        outputs=[ai_msg, ai_chatbot]
+    )
+    ai_chatbot_question_2.click(
+        chat_with_ai,
+        inputs=ai_chatbot_question_2_chat_with_ai_input,
+        outputs=[ai_msg, ai_chatbot]
+    )
+    ai_chatbot_question_3.click(
+        chat_with_ai,
+        inputs=ai_chatbot_question_3_chat_with_ai_input,
+        outputs=[ai_msg, ai_chatbot]
+    )
+
     # file_upload.change(process_file, inputs=file_upload, outputs=df_string_output)
-    file_upload.change(process_file, inputs=file_upload, outputs=[btn_1, btn_2, btn_3, df_summarise, df_string_output])
+    # file_upload.change(process_file, inputs=file_upload, outputs=[btn_1, btn_2, btn_3, df_summarise, df_string_output])
 
     # 当输入 YouTube 链接时触发
-    process_youtube_link_output = [
+    process_youtube_link_inputs = [password, youtube_link]
+    process_youtube_link_outputs = [
         video_id,
         btn_1, 
         btn_2, 
@@ -2154,17 +2288,44 @@ with gr.Blocks(theme=gr.themes.Base(primary_hue=gr.themes.colors.orange, seconda
         content_subject,
         content_grade,
     ]
+    update_state_inputs = [
+        content_subject,
+        content_grade,
+        df_string_output,
+        btn_1, 
+        btn_2, 
+        btn_3
+    ]
+    update_state_outputs = [
+        content_subject_state, 
+        content_grade_state, 
+        trascript_state, 
+        streaming_chat_thread_id_state,
+        ai_chatbot_question_1,
+        ai_chatbot_question_2,
+        ai_chatbot_question_3
+    ]
+
     youtube_link.change(
         process_youtube_link, 
-        inputs=[password,youtube_link], 
-        outputs=process_youtube_link_output
+        inputs=process_youtube_link_inputs, 
+        outputs=process_youtube_link_outputs
+    ).then(
+        update_state,
+        inputs=update_state_inputs,
+        outputs=update_state_outputs
     )
-    
+
     youtube_link_btn.click(
         process_youtube_link, 
-        inputs=[password, youtube_link], 
-        outputs=process_youtube_link_output
+        inputs=process_youtube_link_inputs, 
+        outputs=process_youtube_link_outputs
+    ).then(
+        update_state,
+        inputs=update_state_inputs,
+        outputs=update_state_outputs
     )
+    
 
     # 当输入网页链接时触发
     # web_link.change(process_web_link, inputs=web_link, outputs=[btn_1, btn_2, btn_3, df_summarise, df_string_output])

chatbot.py

@@ -122,7 +122,7 @@ class Chatbot:
         request_payload = {
             "model": "mixtral-8x7b-32768",
             "messages": messages,
-            "max_tokens": 1000  # 設定一個較大的值，可根據需要調整
+            "max_tokens": 500  # 設定一個較大的值，可根據需要調整
         }
         groq_client = self.ai_client
         response = groq_client.chat.completions.create(**request_payload)
@@ -141,7 +141,7 @@ class Chatbot:
             "accept": "application/json",
             "body": json.dumps({
                 "anthropic_version": "bedrock-2023-05-31",
-                "max_tokens": 1000,
+                "max_tokens": 500,
                 "system": system_prompt,
                 "messages": messages
             })

requirements.txt

@@ -1,6 +1,6 @@
-gradio>=4.24.0
+gradio==4.8.0
 pandas
-openai>=1.0.0
+openai>=1.16.2
 requests
 beautifulsoup4
 python-docx