fixed app.py

app.py

@@ -7,17 +7,35 @@ from retriever import (
 )
 from sentence_transformers import CrossEncoder
 
-api_key = os.getenv("OPENROUTER_API_KEY")
-client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=api_key)
+QWEN_MODEL="qwen/qwen3-235b-a22b:free"
+#api_key = os.getenv("OPENROUTER_API_KEY")
+#client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=api_key)
+
+# open ai method
+api_key = os.getenv("OPENAI_API_KEY")
+client = OpenAI(api_key=api_key)
+GPT_MODEL="gpt-4o"
+
 
 collection = load_collection()
 encoder = load_encoder()
 reranker = CrossEncoder("BAAI/bge-reranker-large")
 
+def reformulate_query(user_question, model_name=GPT_MODEL):
+    prompt = f"""你是一个BangDream知识检索助手。请把用户的问题扩写或转写为适合知识库语义检索的检索语句，涵盖所有可能的提问方式或同义关键词。
+    用户问题：{user_question}
+    """
+    resp = client.chat.completions.create(
+        model=model_name,
+        messages=[{"role": "user", "content": prompt}],
+        temperature=0.1,
+        max_tokens=4096,
+    )
+    return resp.choices[0].message.content.strip()
 
 def build_rag_prompt(query, context, system_message):
     prompt = f"""{system_message}
-
+你将获得多个独立的资料片段，请充分查阅每一条资料.
 已知资料如下：
 {context}
 
@@ -49,10 +67,24 @@ def respond(
     system_msg = (system_message or default_system_message).strip()
     chat_history = [{"role": "system", "content": system_msg}]
 
+    # reformulate query
+    print("Reformulating...")
+    reformulated_query_text = reformulate_query(message)
+    print(f"[DEBUG] reformulated query: {reformulated_query_text}")
+
+    print("Thinking...\n...")
+    # rerank original query
     query_vec = encode_query(encoder, message)
-    results = retrieve_docs(collection, query_vec, top_k=30)
+    results = retrieve_docs(collection, query_vec, top_k=20)
     reranked = query_rerank(reranker, message, results, top_n=10)
-    deduped = dedup_by_chapter_event(reranked, max_per_group=1)
+
+    # rerank reformulated query
+    reformulated_query_vec = encode_query(encoder, reformulated_query_text)
+    reformulated_results = retrieve_docs(collection, reformulated_query_vec, top_k=20)
+    reformulated_reranked = query_rerank(reranker, reformulated_query_text, reformulated_results, top_n=10)
+
+    total_reranked = reranked + reformulated_reranked
+    deduped = dedup_by_chapter_event(total_reranked, max_per_group=1)
     expanded_results = expand_with_neighbors(deduped[:3], collection)
     context = expanded_results[0][0] if expanded_results else ""
 
@@ -64,7 +96,7 @@ def respond(
 
     response = ""
     stream = client.chat.completions.create(
-        model="qwen/qwen3-235b-a22b:free",
+        model=GPT_MODEL,
         messages=messages,
         temperature=temperature,
         max_tokens=max_tokens,