first commit

app.py

@@ -0,0 +1,141 @@
+import streamlit as st
+import openai
+
+import os
+import outlines
+# =============== 基本設定 ===============
+#openai.api_key = st.secrets["OPENAI_API_KEY"]  # 建議放在 .streamlit/secrets.toml 中
+
+
+def choose_api_key(model_id):
+    api_key = os.environ["TOGETHER_API_KEY"]
+    if( ('o3-' or 'gpt-') in model_id):
+        api_key = os.environ["CHATGPT_API_KEY"]
+    print(f'api_key:{api_key}')    
+    return api_key
+        
+# =============== 呼叫 LLM 摘要函數 ===============
+def call_llm(model_id, content):
+    api_key = choose_api_key(model_id)
+    client = openai.OpenAI(api_key=api_key,base_url="https://api.together.xyz/v1")
+    
+    response = client.chat.completions.create(
+          model = model_id,
+          messages=[{"role": "user", "content": content}],)
+    return response.choices[0].message.content
+    '''
+
+    response = openai.ChatCompletion.create(
+        #model=model_engine,
+        messages=[{"role": "user", "content": content}],
+        temperature=0.3
+    )'''
+    return response['choices'][0]['message']['content']
+
+# =============== Streamlit 介面 ===============
+st.title("📑 會議紀錄摘要與評估系統")
+
+# 1️⃣ 上傳會議紀錄文字檔
+uploaded_file = st.file_uploader("請上傳會議紀錄文字檔 (.txt)", type="txt")
+
+# 2️⃣ 選擇摘要用的 LLM 模型
+llm_models = [  "Qwen/Qwen2.5-Coder-32B-Instruct", 
+                "Qwen/Qwen2-VL-72B-Instruct", 
+                "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
+                "gpt-4o",
+                "o3-mini",
+                "gpt-3.5-turbo"
+                ]
+selected_summary_model = st.selectbox("選擇用來摘要的 LLM 模型", llm_models)
+
+
+# 3️⃣ 按鈕 → 產生摘要
+if uploaded_file and st.button("🚀 產生會議摘要"):
+    meeting_text = uploaded_file.read().decode("utf-8")
+    if(meeting_text):
+        #summary_task = "請將以下會議紀錄內容進行摘要，並以條列式清楚表達重點："
+        prompt_summary = f'''
+                You are a helpful assistant specialized in summarizing business meeting transcripts. Given a detailed company meeting transcript, your task is to generate a well-structured summary that includes the following:
+                (1)Meeting Outline
+                A high-level bullet-point outline of the meeting topics discussed, in the order they appeared.
+                (2)Speaker Contributions
+                A section listing each participant’s key opinions, statements, and concerns raised during the meeting.
+                (3)Key Conclusions and Action Items
+                A clear list of the main conclusions reached by the team and the action plans assigned to each person or department.
+                
+                The summary should be concise, clear, and organized, avoiding unnecessary repetition while preserving essential details. Your summary should be written in traditional Chinese.
+                
+                ==== 
+                Following is the  meeting transcripts:
+                {meeting_text}
+                '''
+                
+        #cprint(prompt,"yellow")        
+        summary_result = call_llm(selected_summary_model, meeting_text)
+
+        # 存檔
+        summary_filename = f"summary_{selected_summary_model}.txt"
+        with open(summary_filename, "w", encoding="utf-8") as f:
+            f.write(summary_result)
+
+        st.success(f"✅ 已完成摘要，檔名：{summary_filename}")
+        st.text_area("📋 產生的會議摘要", summary_result, height=300)
+
+        # 保留原文與摘要供評估用
+        st.session_state['meeting_text'] = meeting_text
+        st.session_state['summary_result'] = summary_result
+
+# 4️⃣ 選擇評價用的 LLM 模型
+if 'summary_result' in st.session_state:
+    selected_eval_model = st.selectbox("選擇用來評價的 LLM 模型", llm_models)
+
+    # 5️⃣ 按鈕 → 產生評價
+    if st.button("📝 產生摘要評價"):
+        #eval_task = "請根據提供的會議紀錄原文與產生的摘要，評估該摘要的完整性、準確性、表達清晰度，並給予建議。"
+
+        #combined_content = f"【會議紀錄原文】\n{st.session_state['meeting_text']}\n\n【會議摘要】\n{st.session_state['summary_result']}"
+        
+        prompt_evaluation = f'''
+            You are a professional meeting summary reviewer. Please evaluate the following meeting summary based on these three criteria:
+
+            1. Coverage Completeness: Does the summary cover all the key points discussed in the meeting?
+            2. Accuracy: Is the summary faithful to the original meeting content, without adding, omitting, or distorting information?
+            3. Clarity and Conciseness: Is the summary clear, well-structured, and avoids unnecessary repetition or irrelevant content?
+
+            Provide a score from 1 to 5 for each item.  
+            Then, calculate a total score (maximum 15 points).  
+            Finally, write a brief comment about the overall quality.
+
+            ====
+            【會議原文】
+            {st.session_state['meeting_text']}
+
+            【會議摘要】
+            {st.session_state['summary_result']}
+
+            ====
+            Please reply in the following format in Traditional Chinese:
+
+            完整性：X/5  
+            準確性：X/5  
+            簡潔性：X/5  
+            總分：X/15  
+            評語：XXXXX
+            '''
+        
+        evaluation_result = call_llm(selected_eval_model, prompt_evaluation)
+        eval_filename = ""
+        # 存檔
+        '''
+        if(len(selected_eval_model.split('/')) > 1):
+            eval_filename = selected_eval_model.split('/')[1]
+        else:
+            eval_filename = selected_eval_model
+        '''
+        eval_filename = f"evaluation_{eval_filename}.txt"
+        with open(eval_filename, "w", encoding="utf-8") as f:
+            f.write(evaluation_result)
+
+        st.success(f"✅ 已完成評價，檔名：{eval_filename}")
+        st.text_area("📖 評估結果", evaluation_result, height=300)
+

requirements.txt

@@ -0,0 +1,3 @@
+streamlit
+openai
+outlines