Update app.py

app.py

@@ -6,41 +6,50 @@ import time
 from optimum.intel import OVModelForCausalLM, OVWeightQuantizationConfig
 import nncf
 from llama_index.core import SimpleDirectoryReader
-from llama_index.readers.web import BeautifulSoupWebReader
 from rank_bm25 import BM25Okapi
 import jieba
+import requests
+from bs4 import BeautifulSoup
+import os
 
 import huggingface_hub as hf_hub
+
+# 先下載網頁並存成 .txt
+os.makedirs("./data", exist_ok=True)
+urls = [
+    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000311.html",
+    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000305.html",
+    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000306.html",
+    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000312.html",
+]
+for i, url in enumerate(urls):
+    resp = requests.get(url)
+    soup = BeautifulSoup(resp.text, "html.parser")
+    text = soup.get_text(separator="\n", strip=True)
+    with open(f"./data/doc_{i}.txt", "w", encoding="utf-8") as f:
+        f.write(text)
+
 # 初始化 OpenVINO 模型
-#model_id = "hsuwill000/BitCPM4-1B_int4_ov"
 model_id = "hsuwill000/MiniCPM3-4B_int4_ov"
 model_path = "ov"
 hf_hub.snapshot_download(model_id, local_dir=model_path)
 
 config = ov_genai.GenerationConfig()
 config.max_new_tokens = 4096
-config.top_p = 0.9;
-config.top_k = 30;
+config.top_p = 0.9
+config.top_k = 30
 
-reader = BeautifulSoupWebReader()
 pipe = ov_genai.LLMPipeline(model_path, "CPU")
 pipe.get_tokenizer().set_chat_template(pipe.get_tokenizer().chat_template)
 
-
-documents = reader.load_data([
-    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000311.html",
-    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000305.html",
-    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000306.html",
-    "https://www.cwa.gov.tw/V8/C/M/Fishery/tide_30day_MOD/T000312.html",
-])
-
+# 讀取剛剛存的 txt 檔案
+documents = SimpleDirectoryReader("./data").load_data()
 texts = [doc.get_content() for doc in documents]
 
 # 使用 jieba 斷詞做 BM25
 tokenized_corpus = [list(jieba.cut(text)) for text in texts]
 bm25 = BM25Okapi(tokenized_corpus)
 
-
 def start_chat():
     pipe.start_chat()
     return "✅ 開始對話！"
@@ -49,44 +58,36 @@ def finish_chat():
     pipe.finish_chat()
     return "🛑 結束對話！"
 
-
-	
-# 建立推論函式：使用 streamer 並回傳 generator 結果
 def generate_stream(prompt):
-    prompt = prompt #+ " /no_think" + " 答案短且明瞭"
     tokenized_query = list(jieba.cut(prompt))
-    
-    # BM25 取得 top 3 相關文件段落
     top_k = 1
     doc_scores = bm25.get_scores(tokenized_query)
     top_k_indices = sorted(range(len(doc_scores)), key=lambda i: doc_scores[i], reverse=True)[:top_k]
-    
     retrieved_texts = [texts[i] for i in top_k_indices]
-    
+
     print("=== 檢索到的相關段落 ===")
     for i, txt in enumerate(retrieved_texts, 1):
         print(f"--- 段落 {i} ---\n{txt}\n")
-    
-    # 拼接 prompt，避免全文貼上，只用 top3 段落
+
     context = "\n\n".join(retrieved_texts)
     final_prompt = f"根據以下資訊，請簡潔回答問題：\n{context}\n\n問題：{prompt}\n回答："
-    
+
     print("=== 最終 prompt ===")
     print(final_prompt)
-    
+
     q = queue.Queue()
     tps_result = ""
+
     def streamer(subword):
         print(subword, end='', flush=True)
         q.put(subword)
         return ov_genai.StreamingStatus.RUNNING
 
     def worker():
-        # 在背景 thread 中做推論
         nonlocal tps_result
         gen_result = pipe.generate([final_prompt], streamer=streamer, config=config)
         tps = gen_result.perf_metrics.get_throughput().mean
-        tps_result = f"{tps:.2f} tokens/s"        
+        tps_result = f"{tps:.2f} tokens/s"
         q.put(None)  # 結束符號
 
     threading.Thread(target=worker).start()
@@ -97,7 +98,7 @@ def generate_stream(prompt):
         if token is None:
             break
         result += token
-        yield result,""  # 把逐步結果傳給 output textbox
+        yield result, ""
     yield result, tps_result
 
 with gr.Blocks() as demo:
@@ -111,13 +112,12 @@ with gr.Blocks() as demo:
     with gr.Row():
         textbox_input = gr.Textbox(label="Prompt", lines=1, placeholder="Enter prompt here...")
         button = gr.Button("Submit")
-        
+
     textbox_output = gr.Markdown(label="robot answer:", elem_id="scroll_output")
 
     start_btn.click(fn=start_chat, outputs=status_box)
-    end_btn.click(fn=finish_chat, outputs=status_box)   
+    end_btn.click(fn=finish_chat, outputs=status_box)
 
-    # 當按鈕被按下時，呼叫 generate_stream 並更新 textbox_output
     button.click(fn=generate_stream, inputs=textbox_input, outputs=[textbox_output, TPS_box])
 
-demo.launch()
+demo.launch()