save code

services/llm.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import logging
 import os
+import re
 from typing import Optional
 from collections import deque
 
@@ -18,7 +19,7 @@ class LLMService:
         self._model = (settings.llm_model or "").strip()
         self._base_url = (settings.llm_base_url or "").strip()
         self._history: deque[dict] = deque(maxlen=12)  # 6 turns (user+assistant)
-        self._system_prompt = "请用中文口语化简短回复，1-2句。"
+        self._system_prompt = (settings.llm_system_prompt or "").strip() or "请用中文口语化简短回复，1-2句。"
 
         if not settings.llm_api_key:
             self._error = "missing LLM api key: set LLM_API_KEY or DEEPSEEK_API_KEY"
@@ -51,6 +52,22 @@ class LLMService:
             self._error = str(exc)
             logger.warning("LLM online mode unavailable: %s", exc)
 
+    @staticmethod
+    def _normalize_reply_text(text: str) -> str:
+        cleaned = (text or "").strip()
+        if not cleaned:
+            return ""
+        cleaned = cleaned.replace("\r", "\n")
+        cleaned = re.sub(r"```.*?```", "", cleaned, flags=re.S)
+        cleaned = re.sub(r"`([^`]*)`", r"\1", cleaned)
+        cleaned = re.sub(r"\*\*(.*?)\*\*", r"\1", cleaned)
+        cleaned = re.sub(r"\*(.*?)\*", r"\1", cleaned)
+        cleaned = re.sub(r"\s*([，。！？；：、,.!?;:])\s*", r"\1", cleaned)
+        cleaned = re.sub(r"(?<=[\u4e00-\u9fff])\s+(?=[\u4e00-\u9fff])", "", cleaned)
+        cleaned = re.sub(r"\n+", " ", cleaned)
+        cleaned = re.sub(r"\s{2,}", " ", cleaned)
+        return cleaned.strip()
+
     async def reply_with_meta(self, text: str) -> tuple[str, str]:
         if not self._ready or self._client is None:
             return f"收到：{text}。这是本地兜底回复。", "fallback"
@@ -60,9 +77,9 @@ class LLMService:
                 model=self._model,
                 messages=[{"role": "system", "content": self._system_prompt}, *list(self._history), {"role": "user", "content": text}],
                 max_tokens=settings.llm_max_tokens,
-                temperature=0.7,
+                temperature=max(0.1, min(1.2, settings.llm_temperature)),
             )
-            content = (resp.choices[0].message.content or "").strip()
+            content = self._normalize_reply_text(resp.choices[0].message.content or "")
             if content:
                 self._history.append({"role": "user", "content": text})
                 self._history.append({"role": "assistant", "content": content})