Update app.py

app.py

@@ -1,9 +1,11 @@
 import os
 import sys
 import math
+from openai import OpenAI
 import requests
 import gradio as gr
 import pandas as pd
+import concurrent.futures
 from datasets import Dataset
 from tqdm import tqdm
 from ragas import evaluate, SingleTurnSample
@@ -18,7 +20,7 @@ from ragas.metrics import (
 # 設定輸出編碼為 UTF-8（解決中文顯示問題）
 sys.stdout.reconfigure(encoding="utf-8")
 
-# 支援從Google Drive下載 Ground Truth
+# 從Google Drive下載 Ground Truth
 gt_url = os.environ.get("GT_URL")
 gt_path = "tender_groundtruth.csv"
 
@@ -55,8 +57,58 @@ def log_to_google_sheet(question, answer, contexts, scores):
     except Exception as e:
         print("寫入 Google Sheet 失敗:", str(e))
 
+def fetch_sheet_content():
+    DEFAULT_ANNOUNCEMENT = "尚無公告"
+    DEFAULT_FAQ = "尚無常見問題"
+
+    try:
+        url = os.environ.get("ANNOUNCEMENT_URL")
+        if not url:
+            print("Warning: 未設定 ANNOUNCEMENT_URL")
+            return DEFAULT_ANNOUNCEMENT, DEFAULT_FAQ
+
+        df = pd.read_csv(url)
+
+        announcement = df["Announcement"].iloc[0].strip() if "Announcement" in df.columns else DEFAULT_ANNOUNCEMENT
+        faq = df["FAQ"].iloc[0].strip() if "FAQ" in df.columns else DEFAULT_FAQ
+
+        announcement = announcement.replace("\\n", "<br>").replace("\n", "<br>")
+        faq = faq.replace("\\n", "<br>").replace("\n", "<br>")
+
+        return announcement or DEFAULT_ANNOUNCEMENT, faq or DEFAULT_FAQ
+
+    except Exception as e:
+        print("載入 Sheet 錯誤：", e)
+        return DEFAULT_ANNOUNCEMENT, DEFAULT_FAQ
+
+
+def validate_openai_key(api_key):
+    try:
+        client = OpenAI(api_key=api_key)
+        client.chat.completions.create(
+            model="gpt-3.5-turbo",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=1
+        )
+        return None
+    except Exception as e:
+        err_msg = str(e)
+        if "Incorrect API key provided" in err_msg:
+            return pd.DataFrame([{"錯誤訊息": " 您輸入的 OpenAI API Key 有誤，請確認是否貼錯、字數不符或格式異常。"}]), None
+        elif "exceeded your current quota" in err_msg:
+            return pd.DataFrame([{"錯誤訊息": "您的 OpenAI 帳戶額度已用盡，請前往帳戶頁面檢查餘額。"}]), None
+        elif "Rate limit" in err_msg:
+            return pd.DataFrame([{"錯誤訊息": "OpenAI 請求頻率過高，請稍後再試"}]), None
+        else:
+            return pd.DataFrame([{"錯誤訊息": f"API Key 錯誤：{err_msg}"}]), None
+    
 def RAG_evaluation(uploaded_file, user_api_key):
     try:
+        # 檢查 OpenAI API Key 是否有效
+        validation_result = validate_openai_key(user_api_key)
+        if validation_result:
+            return validation_result
+        
         os.environ["OPENAI_API_KEY"] = user_api_key
         print("評估開始")
 
@@ -76,10 +128,10 @@ def RAG_evaluation(uploaded_file, user_api_key):
             print("未合併題目：", missing["Question"].tolist())
         if merged_df.empty:
             return pd.DataFrame([{"錯誤訊息": "合併後無資料，請確認題目與 GT 是否對應"}]), None
-
+        
         llm_wrapper = LangchainLLMWrapper(ChatOpenAI(model="gpt-4o-mini-2024-07-18"))
         embedding_wrapper = LangchainEmbeddingsWrapper(OpenAIEmbeddings(model="text-embedding-3-large"))
-
+        
         batch_size = 10
         records = []
         for batch_start in tqdm(range(0, len(merged_df), batch_size), desc="RAGAS Batch Evaluating"):
@@ -104,13 +156,18 @@ def RAG_evaluation(uploaded_file, user_api_key):
                 result = evaluate(
                     dataset=dataset,
                     metrics=[
-                        LLMContextPrecisionWithReference(), LLMContextRecall(), ContextEntityRecall(),
-                        NoiseSensitivity(), Faithfulness(), ResponseRelevancy(),
-                        SemanticSimilarity(), FactualCorrectness()
+                        LLMContextPrecisionWithReference(),     # context precision
+                        LLMContextRecall(),                     # context recall     
+                        ContextEntityRecall(),
+                        # NoiseSensitivity(), 
+                        Faithfulness(),                         # faithfulness
+                        ResponseRelevancy(),                    # answer relevancy
+                        SemanticSimilarity(),                   # semantic similarity
+                        # FactualCorrectness()
                     ],
                     llm=llm_wrapper,
                     embeddings=embedding_wrapper,
-                    show_progress=False
+                    show_progress=True
                 )
 
                 result_df = result.to_pandas()
@@ -128,7 +185,7 @@ def RAG_evaluation(uploaded_file, user_api_key):
                         "Context Entity Recall": getattr(row, "context_entity_recall", None),
                         # "Noise Sensitivity": getattr(row, "noise_sensitivity_relevant", None)
                     }
-                    
+
                     for key in list(record.keys()):
                         val = record[key]
                         if isinstance(val, float) and not math.isfinite(val):
@@ -156,9 +213,11 @@ def RAG_evaluation(uploaded_file, user_api_key):
             avg_row["Question"] = "Average"
             score_df = pd.concat([score_df, pd.DataFrame([avg_row])], ignore_index=True)
 
-        output_path = "result_output.csv"
+        original_name = os.path.basename(uploaded_file.name)
+        filename = os.path.splitext(original_name)[0]
+        output_path = f"{filename}_result.csv"
         score_df.to_csv(output_path, index=False, encoding="utf-8-sig")
-        print("評估結果已儲存為 CSV：", output_path)
+        print("評估結果已儲存：", output_path)
 
         return score_df, output_path
 
@@ -192,30 +251,45 @@ def check_csv_and_run(file, key):
             missing_questions = "\n".join(f"- {q}" for q in invalid_rows["Question"].tolist())
             return pd.DataFrame([{"錯誤訊息": f"發現 {len(invalid_rows)} 筆資料中 Answer 或 Context 為空：\n{missing_questions}"}]), None
         
+        # check eval context
         try:
             for i, val in df["Context"].dropna().items():
                 if not isinstance(eval(val), list):
                     return pd.DataFrame([{"錯誤訊息": f"第 {i + 1} 筆 Context 欄格式錯誤，請確認其內容應為 list"}]), None
         except Exception as e:
             return pd.DataFrame([{"錯誤訊息": f"Context 欄格式解析錯誤，請確認其為有效的 list 格式，例如 ['A', 'B']：{str(e)}"}]), None
-
-    except Exception as e:
-        return pd.DataFrame([{"錯誤訊息": f"發生錯誤：{str(e)}"}]), None
-
-    # 若上傳之待評估檔案無錯誤，執行評估
-    try:
-        return RAG_evaluation(file, key)
+        
+        # 若上傳之待評估檔案無錯誤，執行評估
+        try:
+            return RAG_evaluation(file, key)
+        # 檢查 OpenAI API Key 是否有效
+        except Exception as e:
+            error_message = str(e)
+            return pd.DataFrame([{"錯誤訊息": f"系統錯誤：{error_message}"}]), None
     except Exception as e:
-        return pd.DataFrame([{"錯誤訊息": f"RAG 評估失敗：{str(e)}"}]), None
+        return pd.DataFrame([{"錯誤訊息": f"評估失敗：{str(e)}"}]), None
 
 # Gradio 介面
 with gr.Blocks() as demo:
-    gr.Markdown("## 📐 RAG系統評估工具")
     gr.Markdown("""
-    ### 📄 使用說明
-    請上傳您RAG系統產出的結果檔案（包含 Question, Context, Answer 欄位），並填入您的OpenAI API Key，以評估您的RAG系統。  
-    #### ⏳ 完整評估需要數小時，無即時回應並不是當機，請耐心等候。
-    """)
+# 📐 RAG系統評估工具 (分流B)
+                
+### 📄 使用說明
+- 請上傳您 RAG 系統產出的結果檔案（需包含欄位：Question、Context、Answer），並填入您的 OpenAI API Key，以進行評估。
+- ⏳ 完整評估**通常需耗時 1 小時以上**，若無即時回應，請**耐心等候**，系統並未當機，謝謝您的理解。
+
+### 🚦 分流措施
+本工具部署於 Hugging Face Public Space，若同時有多位使用者使用，系統會將您的評估請求**排入佇列**。
+為避免長時間等待，建議您**先僅送出 1 筆資料進行測試**，若進度條顯示之預估**等待時間超過 2 小時（7000 秒以上），可能是其他使用者正在使用**。
+
+本頁為**分流 B**，您可以考慮改用其他分流或稍後再試，感謝您的耐心與配合!
+- 🔁 [分流 A](https://huggingface.co/spaces/KSLab/RAG_Evaluator_A)
+- 🔁 [分流 B](https://huggingface.co/spaces/KSLab/RAG_Evaluator_B)
+- 🔁 [分流 C](https://huggingface.co/spaces/KSLab/RAG_Evaluator_C)
+
+### 📢 系統公告
+""")
+    announcement_display = gr.Markdown()
 
     file_input = gr.File(label="上傳 Evaluation_Dataset.csv")
     api_key_input = gr.Textbox(label="OpenAI API Key", type="password")
@@ -224,13 +298,25 @@ with gr.Blocks() as demo:
     result_output = gr.Dataframe(label="評估結果")
     download_link = gr.File(label="下載評估結果(CSV)")
 
+    # 常見QA
+    gr.Markdown("""
+    ---  
+    ### ❓ 常見問題 & 解答
+    """)
+    faq_display = gr.Markdown()
+
+    # 載入公告與 FAQ
+    def load_sheet():
+        return fetch_sheet_content()
+    demo.load(fn=load_sheet, inputs=[], outputs=[announcement_display, faq_display])
+
     def wrapped_fn(file, key):
         return RAG_evaluation(file, key)
 
     submit_btn.click(
         fn=check_csv_and_run,
         inputs=[file_input, api_key_input],
-        outputs=[result_output, download_link]
+        outputs=[result_output, download_link],
 )
 
 demo.launch()