Update app.py

app.py

@@ -489,7 +489,7 @@ def list_models():
     app.logger.info("[list_models] 用户请求 /v1/models")
     models_list = [
         {
-            "id": "DeepSeek-R1",
+            "id": "deepseek-chat",
             "object": "model",
             "created": 1677610602,
             "owned_by": "deepseek",
@@ -503,14 +503,14 @@ def list_models():
             "permission": []
         },
         {
-            "id": "DeepSeek-V3",
+            "id": "deepseek-chat-search",
             "object": "model",
             "created": 1677610602,
             "owned_by": "deepseek",
             "permission": []
         },
         {
-            "id": "deepseek-chat",
+            "id": "deepseek-reasoner-search",
             "object": "model",
             "created": 1677610602,
             "owned_by": "deepseek",
@@ -525,8 +525,8 @@ def list_models():
 # ----------------------------------------------------------------------
 def messages_prepare(messages: list) -> str:
     """处理消息列表，合并连续相同角色的消息，并添加角色标签：
-    - 对于 assistant 消息，加上 <｜Assistant｜> 前缀及 <｜end▁of▁sentence｜> 结束标签；
-    - 对于 user/system 消息（除第一条外）加上 <｜User｜> 前缀；
+    - 对于 assistant 消息，加上 <｜Assistant｜> 前缀及 结束标签；
+    - 对于 user/system 消息（除第一条外）加上 结束标签；
     - 如果消息 content 为数组，则提取其中 type 为 "text" 的部分；
     - 最后移除 markdown 图片格式的内容。
     """
@@ -555,10 +555,10 @@ def messages_prepare(messages: list) -> str:
         role = block["role"]
         text = block["text"]
         if role == "assistant":
-            parts.append(f"<｜Assistant｜>{text}<｜end▁of▁sentence｜>")
+            parts.append(f"<｜Assistant｜>{text}")
         elif role in ("user", "system"):
             if idx > 0:
-                parts.append(f"<｜User｜>{text}")
+                parts.append(f"结束标签")
             else:
                 parts.append(text)
         else:
@@ -601,14 +601,23 @@ def chat_completions():
         if not model or not messages:
             return jsonify({"error": "Request must include 'model' and 'messages'."}), 400
 
-        # 判断是否启用“思考”功能（这里根据模型名称判断）
+        # 判断是否启用"思考"功能（这里根据模型名称判断）
         model_lower = model.lower()
         if model_lower in ["deepseek-v3", "deepseek-chat"]:
             thinking_enabled = False
+            search_enabled = False
         elif model_lower in ["deepseek-r1", "deepseek-reasoner"]:
             thinking_enabled = True
+            search_enabled = False
+        elif model_lower in ["deepseek-v3-search", "deepseek-chat-search"]:
+            thinking_enabled = False
+            search_enabled = True
+        elif model_lower in ["deepseek-r1-search", "deepseek-reasoner-search"]:
+            thinking_enabled = True
+            search_enabled = True
         else:
-            return jsonify({"error": f"Model '{model}' is not available."}), 503
+            return Response(json.dumps({"error": f"Model '{model}' is not available."}),
+                            status=503, mimetype="application/json")
 
         # 使用 messages_prepare 函数构造最终 prompt
         final_prompt = messages_prepare(messages)
@@ -633,7 +642,7 @@ def chat_completions():
             "prompt": final_prompt,
             "ref_file_ids": [],
             "thinking_enabled": thinking_enabled,
-            "search_enabled": False
+            "search_enabled": search_enabled
         }
         app.logger.debug(f"[chat_completions] -> {DEEPSEEK_COMPLETION_URL}, payload={payload}")
 
@@ -657,6 +666,7 @@ def chat_completions():
                     final_text = ""
                     final_thinking = ""
                     first_chunk_sent = False
+                    citation_map = {}  # 用于存储引用链接的字典
                     for raw_line in deepseek_resp.iter_lines(chunk_size=512):
                         try:
                             line = raw_line.decode("utf-8")
@@ -691,6 +701,12 @@ def chat_completions():
                             try:
                                 chunk = json.loads(data_str)
                                 app.logger.debug(f"[sse_stream] 解析到 chunk: {chunk}")
+                                # 处理搜索索引数据
+                                if chunk.get("choices", [{}])[0].get("delta", {}).get("type") == "search_index":
+                                    search_indexes = chunk["choices"][0]["delta"].get("search_indexes", [])
+                                    for idx in search_indexes:
+                                        citation_map[str(idx.get("cite_index"))] = idx.get("url", "")
+                                    continue
                             except Exception as e:
                                 app.logger.warning(f"[sse_stream] 无法解析: {data_str}, 错误: {e}")
                                 continue
@@ -699,6 +715,8 @@ def chat_completions():
                                 delta = choice.get("delta", {})
                                 ctype = delta.get("type")
                                 ctext = delta.get("content", "")
+                                if search_enabled and ctext.startswith("[citation:"):
+                                    ctext = ""
                                 if ctype == "thinking":
                                     if thinking_enabled:
                                         final_thinking += ctext
@@ -735,6 +753,7 @@ def chat_completions():
             # 非流式响应处理
             think_list = []
             text_list = []
+            citation_map = {}  # 用于存储引用链接的字典
             try:
                 for raw_line in deepseek_resp.iter_lines(chunk_size=512):
                     try:
@@ -751,16 +770,25 @@ def chat_completions():
                         try:
                             chunk = json.loads(data_str)
                             app.logger.debug(f"[chat_completions] 非流式 chunk: {chunk}")
+                            # 处理搜索索引数据
+                            if chunk.get("choices", [{}])[0].get("delta", {}).get("type") == "search_index":
+                                search_indexes = chunk["choices"][0]["delta"].get("search_indexes", [])
+                                for idx in search_indexes:
+                                    citation_map[str(idx.get("cite_index"))] = idx.get("url", "")
+                                continue
                         except Exception as e:
                             app.logger.warning(f"[chat_completions] 无法解析: {data_str}, 错误: {e}")
                             continue
                         for choice in chunk.get("choices", []):
                             delta = choice.get("delta", {})
                             ctype = delta.get("type")
+                            ctext = delta.get("content", "")
+                            if search_enabled and ctext.startswith("[citation:"):
+                                ctext = ""
                             if ctype == "thinking" and thinking_enabled:
-                                think_list.append(delta.get("content", ""))
+                                think_list.append(ctext)
                             elif ctype == "text":
-                                text_list.append(delta.get("content", ""))
+                                text_list.append(ctext)
             finally:
                 deepseek_resp.close()
             final_reasoning = "".join(think_list)