feat: 添加深度思考适配

server/utils/glm_adapter.py

@@ -196,7 +196,6 @@ def build_glm_messages(messages: list, files: list | None = None) -> tuple[list,
 _SENTINEL = object()
 
 
-# ── 流式 SSE 生成器 ───────────────────────────────────────────────────
 async def glm_stream_generator(
     messages: list,
     model: str,
@@ -255,6 +254,9 @@ async def glm_stream_generator(
 
     loop = asyncio.get_running_loop()
 
+    full_reasoning = ""   # 累计思考内容（用于判断是否首次）
+    full_content   = ""   # 累计正式回答（用于判断是否首次）
+
     while True:
         item = await loop.run_in_executor(None, chunk_queue.get)
 
@@ -268,17 +270,40 @@ async def glm_stream_generator(
 
         try:
             delta = item.choices[0].delta
-            text = getattr(delta, "content", "") or ""
-            if not text:
+            reasoning = getattr(delta, "reasoning_content", "") or ""
+            text      = getattr(delta, "content", "")      or ""
+
+            delta_str = ""
+
+            # ── 思考过程（reasoning_content）────────────────────────
+            if reasoning:
+                if not full_reasoning:
+                    # 首个思考片段：加 Markdown 引用块标题
+                    delta_str += "> **💭 深度思考过程：**\n> \n> "
+                full_reasoning += reasoning
+                # 引用块内换行需在每行前加 `> `
+                delta_str += reasoning.replace("\n", "\n> ")
+
+            # ── 正式回答（content）──────────────────────────────────
+            if text:
+                if not full_content and full_reasoning:
+                    # 思考结束后首次出现正式回答：加分隔线
+                    delta_str += "\n\n---\n\n"
+                full_content += text
+                delta_str += text
+
+            if not delta_str:
                 continue
+
             data = {
                 "id": f"chatcmpl-{generate_unique_id()}",
                 "object": "chat.completion.chunk",
                 "created": get_current_timestamp(),
                 "model": actual_model,
-                "choices": [{"index": 0, "delta": {"content": text}, "finish_reason": None}],
+                "choices": [{"index": 0, "delta": {"content": delta_str}, "finish_reason": None}],
             }
             yield f"data: {json.dumps(data, ensure_ascii=False)}\n\n"
+
         except Exception as e:
             print(f"[GLM] chunk 解析异常：{e}")
 
@@ -293,6 +318,7 @@ async def glm_stream_generator(
     yield "data: [DONE]\n\n"
 
 
+
 # ── 非流式调用 ────────────────────────────────────────────────────────
 def glm_chat_sync(
     messages: list,