From ce578f0417fc79d59bb0f9b6f38a17c70465e3e2 Mon Sep 17 00:00:00 2001
From: Soulter <905617992@qq.com>
Date: Wed, 5 Feb 2025 15:40:52 +0800
Subject: [PATCH] =?UTF-8?q?feat:=20=E6=94=AF=E6=8C=81=E4=BD=BF=E7=94=A8=20?=
 =?UTF-8?q?LLM=20=E8=BE=85=E5=8A=A9=E5=88=86=E6=AE=B5=E5=9B=9E=E5=A4=8D=20?=
 =?UTF-8?q?#338?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 astrbot/core/config/default.py                |  6 ++++++
 .../process_stage/method/llm_request.py       |  2 +-
 .../core/pipeline/result_decorate/stage.py    | 19 +++++++++++++++++--
 3 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/astrbot/core/config/default.py b/astrbot/core/config/default.py
index 0f18ba243..ba880ef45 100644
--- a/astrbot/core/config/default.py
+++ b/astrbot/core/config/default.py
@@ -29,6 +29,7 @@ DEFAULT_CONFIG = {
             "enable": False,
             "only_llm_result": True,
             "interval": "1.5,3.5",
+            "seg_prompt": "",
             "regex": ".*?[。？！~…]+|.+$"
         },
         "no_permission_reply": True,
@@ -217,6 +218,11 @@ CONFIG_METADATA_2 = {
                                 "type": "string",
                                 "hint": "每一段回复的间隔时间，格式为 `最小时间,最大时间`。如 `0.75,2.5`",
                             },
+                            "seg_prompt": {
+                                "description": "分段提示词辅助",
+                                "type": "string",
+                                "hint": "此项为空时表达不启用这个方法。此方法会调用一次LLM请求。让 LLM 在某一句话中插入一个可以用正则表达式分隔的标记，来实现LLM基于情感分段。如: `请基于情感对以下文本进行分段, 并在两段之间添加`<seg>`以便我用正则匹配。`    然后将下面的正则表达式更换为`.+?<seg>`。",
+                            },
                             "regex": {
                                 "description": "正则表达式",
                                 "type": "string",
diff --git a/astrbot/core/pipeline/process_stage/method/llm_request.py b/astrbot/core/pipeline/process_stage/method/llm_request.py
index dee23211a..b2ee8d508 100644
--- a/astrbot/core/pipeline/process_stage/method/llm_request.py
+++ b/astrbot/core/pipeline/process_stage/method/llm_request.py
@@ -77,7 +77,7 @@ class LLMRequestSubStage(Stage):
             req.contexts = json.loads(req.contexts)
         
         try:
-            logger.debug(f"提供商请求 Payload: {req.__dict__}")
+            logger.debug(f"提供商请求 Payload: {req}")
             if _nested:
                 req.func_tool = None # 暂时不支持递归工具调用
             llm_response = await provider.text_chat(**req.__dict__) # 请求 LLM
diff --git a/astrbot/core/pipeline/result_decorate/stage.py b/astrbot/core/pipeline/result_decorate/stage.py
index 60720ceb7..332e1cef4 100644
--- a/astrbot/core/pipeline/result_decorate/stage.py
+++ b/astrbot/core/pipeline/result_decorate/stage.py
@@ -23,6 +23,7 @@ class ResultDecorateStage:
         # 分段回复
         self.enable_segmented_reply = ctx.astrbot_config['platform_settings']['segmented_reply']['enable']
         self.only_llm_result = ctx.astrbot_config['platform_settings']['segmented_reply']['only_llm_result']
+        self.seg_prompt = ctx.astrbot_config['platform_settings']['segmented_reply']['seg_prompt']
         self.regex = ctx.astrbot_config['platform_settings']['segmented_reply']['regex']
 
     async def process(self, event: AstrMessageEvent) -> Union[None, AsyncGenerator[None, None]]:
@@ -49,12 +50,26 @@ class ResultDecorateStage:
                     new_chain = []
                     for comp in result.chain:
                         if isinstance(comp, Plain):
-                            split_response = re.findall(r".*?[。？！~…]+|.+$", comp.text)
+                            
+                            if self.seg_prompt:
+                                try:
+                                    llm_resp = await self.ctx.plugin_manager.context.get_using_provider().text_chat(
+                                        prompt=f"{self.seg_prompt}\n{comp.text}",
+                                    )
+                                    comp.text = llm_resp.completion_text
+                                except BaseException as e:
+                                    traceback.print_exc()
+                                    logger.error("使用 LLM 分段回复失败： " + str(e))
+                                    new_chain.append(comp)
+                                    continue
+                            
+                            split_response = re.findall(self.regex, comp.text)
                             if not split_response:
                                 new_chain.append(comp)
                                 continue
                             for seg in split_response:
-                                new_chain.append(Plain(seg))
+                                if seg:
+                                    new_chain.append(Plain(seg))
                         else:
                             # 非 Plain 类型的消息段不分段
                             new_chain.append(comp)