fix:manager意图识别bug (#762)

* fix:连接manager后无法使用functioncallbug

* fix:意图识别使用llm无法播放音乐bug

* fix：manager第一图识别使用独立llm无法初始化llm的bug

main/xiaozhi-server/core/connection.py

@@ -104,8 +104,6 @@ class ConnectionHandler:
 
         self.close_after_chat = False  # 是否在聊天结束后关闭连接
         self.use_function_call_mode = False
-        if self.config["selected_module"]["Intent"] == "function_call":
-            self.use_function_call_mode = True
 
     async def handle_connection(self, ws):
         try:
@@ -222,37 +220,37 @@ class ConnectionHandler:
         )
         if private_config.get("VAD", None) is not None:
             init_vad = True
-            self.config["vad"] = private_config["VAD"]
+            self.config["VAD"] = private_config["VAD"]
             self.config["selected_module"]["VAD"] = private_config["selected_module"][
                 "VAD"
             ]
         if private_config.get("ASR", None) is not None:
             init_asr = True
-            self.config["asr"] = private_config["ASR"]
+            self.config["ASR"] = private_config["ASR"]
             self.config["selected_module"]["ASR"] = private_config["selected_module"][
                 "ASR"
             ]
         if private_config.get("LLM", None) is not None:
             init_llm = True
-            self.config["llm"] = private_config["LLM"]
+            self.config["LLM"] = private_config["LLM"]
             self.config["selected_module"]["LLM"] = private_config["selected_module"][
                 "LLM"
             ]
         if private_config.get("TTS", None) is not None:
             init_tts = True
-            self.config["tts"] = private_config["TTS"]
+            self.config["TTS"] = private_config["TTS"]
             self.config["selected_module"]["TTS"] = private_config["selected_module"][
                 "TTS"
             ]
         if private_config.get("Memory", None) is not None:
             init_memory = True
-            self.config["memory"] = private_config["Memory"]
+            self.config["Memory"] = private_config["Memory"]
             self.config["selected_module"]["Memory"] = private_config[
                 "selected_module"
             ]["Memory"]
         if private_config.get("Intent", None) is not None:
             init_intent = True
-            self.config["intent"] = private_config["Intent"]
+            self.config["Intent"] = private_config["Intent"]
             self.config["selected_module"]["Intent"] = private_config[
                 "selected_module"
             ]["Intent"]
@@ -287,17 +285,26 @@ class ConnectionHandler:
         self.memory.init_memory(device_id, self.llm)
 
     def _initialize_intent(self):
+        if (
+            self.config["Intent"][self.config["selected_module"]["Intent"]]["type"]
+            == "function_call"
+        ):
+            self.use_function_call_mode = True
         """初始化意图识别模块"""
         # 获取意图识别配置
         intent_config = self.config["Intent"]
-        intent_type = self.config["selected_module"]["Intent"]
+        intent_type = self.config["Intent"][self.config["selected_module"]["Intent"]][
+            "type"
+        ]
 
         # 如果使用 nointent，直接返回
         if intent_type == "nointent":
             return
         # 使用 intent_llm 模式
         elif intent_type == "intent_llm":
-            intent_llm_name = intent_config["intent_llm"]["llm"]
+            intent_llm_name = intent_config[self.config["selected_module"]["Intent"]][
+                "llm"
+            ]
 
             if intent_llm_name and intent_llm_name in self.config["LLM"]:
                 # 如果配置了专用LLM，则创建独立的LLM实例

main/xiaozhi-server/core/handle/functionHandler.py

@@ -57,7 +57,9 @@ class FunctionHandler:
 
     def register_config_functions(self):
         """注册配置中的函数,可以不同客户端使用不同的配置"""
-        for func in self.config["Intent"]["function_call"].get("functions", []):
+        for func in self.config["Intent"][self.config["selected_module"]["Intent"]].get(
+            "functions", []
+        ):
             self.function_registry.register_function(func)
 
         """home assistant需要初始化提示词"""

main/xiaozhi-server/core/handle/intentHandler.py

@@ -76,6 +76,11 @@ async def process_intent_result(conn, intent_result, original_text):
             if function_name == "continue_chat":
                 return False
 
+            if function_name == "play_music":
+                funcItem = conn.func_handler.get_function(function_name)
+                if not funcItem:
+                    conn.func_handler.function_registry.register_function("play_music")
+
             function_args = None
             if "arguments" in intent_data["function_call"]:
                 function_args = intent_data["function_call"]["arguments"]

main/xiaozhi-server/core/handle/iotHandle.py

@@ -269,14 +269,12 @@ def register_device_type(descriptor):
 
 # 用于接受前端设备推送的搜索iot描述
 async def handleIotDescriptors(conn, descriptors):
-    if not conn.use_function_call_mode:
-        return
     wait_max_time = 5
     while conn.func_handler is None or not conn.func_handler.finish_init:
         await asyncio.sleep(1)
         wait_max_time -= 1
         if wait_max_time <= 0:
-            logger.bind(tag=TAG).error("连接对象没有func_handler")
+            logger.bind(tag=TAG).debug("连接对象没有func_handler")
             return
     """处理物联网描述"""
     functions_changed = False

main/xiaozhi-server/core/providers/llm/openai/openai.py

@@ -15,7 +15,15 @@ class LLMProvider(LLMProviderBase):
             self.base_url = config.get("base_url")
         else:
             self.base_url = config.get("url")
-        self.max_tokens = config.get("max_tokens", 500)
+        max_tokens = config.get("max_tokens")
+        if max_tokens is None or max_tokens == "":
+            max_tokens = 500
+
+        try:
+            max_tokens = int(max_tokens)
+        except (ValueError, TypeError):
+            max_tokens = 500
+        self.max_tokens = max_tokens
 
         check_model_key("LLM", self.api_key)
         self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)

main/xiaozhi-server/core/providers/vad/silero.py

@@ -42,7 +42,8 @@ class VADProvider(VADProviderBase):
                 audio_tensor = torch.from_numpy(audio_float32)
 
                 # 检测语音活动
-                speech_prob = self.model(audio_tensor, 16000).item()
+                with torch.no_grad():
+                    speech_prob = self.model(audio_tensor, 16000).item()
                 client_have_voice = speech_prob >= self.vad_threshold
 
                 # 如果之前有声音，但本次没有声音，且与上次有声音的时间查已经超过了静默阈值，则认为已经说完一句话