Merge pull request #440 from xinnan-tech/test-server-pr

Test server pr

main/xiaozhi-server/config.yaml

@@ -142,6 +142,7 @@ VAD:
     min_silence_duration_ms: 700  # 如果说话停顿比较长，可以把这个值设置大一些
 
 LLM:
+  # 所有openai类型均可以修改超参，以AliLLM为例
   # 当前支持的type为openai、dify、ollama，可自行适配
   AliLLM:
     # 定义LLM API类型
@@ -150,6 +151,11 @@ LLM:
     base_url: https://dashscope.aliyuncs.com/compatible-mode/v1
     model_name: qwen-turbo
     api_key: 你的deepseek web key
+    temperature: 0.7  # 温度值
+    max_tokens: 500   # 最大生成token数
+    top_p: 1         
+    top_k: 50        
+    frequency_penalty: 0  # 频率惩罚
   DoubaoLLM:
     # 定义LLM API类型
     type: openai
@@ -198,7 +204,9 @@ LLM:
     # token申请地址： https://aistudio.google.com/apikey
     # 若部署地无法访问接口，需要开启科学上网
     api_key: 你的gemini web key
-    model_name: "gemini-1.5-pro"  #  gemini-1.5-pro 是免费的
+    model_name: "gemini-2.0-flash"
+    http_proxy: ""  #"http://127.0.0.1:10808"
+    https_proxy: "" #http://127.0.0.1:10808"
   CozeLLM:
     # 定义LLM API类型
     type: coze

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

@@ -1,14 +1,19 @@
 import google.generativeai as genai
 from core.utils.util import check_model_key
 from core.providers.llm.base import LLMProviderBase
-
+from config.logger import setup_logging
+import requests
+import json
+TAG = __name__
+logger = setup_logging()
 
 class LLMProvider(LLMProviderBase):
     def __init__(self, config):
         """初始化Gemini LLM Provider"""
         self.model_name = config.get("model_name", "gemini-1.5-pro")
         self.api_key = config.get("api_key")
-
+        self.http_proxy=config.get("http_proxy")
+        self.https_proxy = config.get("https_proxy")
         have_key = check_model_key("LLM", self.api_key)
 
         if not have_key:
@@ -16,6 +21,19 @@ class LLMProvider(LLMProviderBase):
 
         try:
             # 初始化Gemini客户端
+            # 配置代理（如果提供了代理配置）
+            self.proxies=None
+            if self.http_proxy is not "" or self.https_proxy is not "":
+
+                self.proxies = {
+                    "http": self.http_proxy,
+                    "https": self.https_proxy,
+                }
+                logger.bind(tag=TAG).info(f"Gemini set proxys:{self.proxies}")
+                # 使用猴子补丁修改 google-generativeai 库的请求会话
+
+                # 使用 session 对象配置 genai
+
             genai.configure(api_key=self.api_key)
             self.model = genai.GenerativeModel(self.model_name)
 
@@ -46,26 +64,54 @@ class LLMProvider(LLMProviderBase):
                 if content:
                     chat_history.append({
                         "role": role,
-                        "parts": [content]
+                        "parts": [{"text":content}]
+
                     })
 
             # 获取当前消息
             current_msg = dialogue[-1]["content"]
 
-            # 创建新的聊天会话
+            # 构建请求体
-            chat = self.model.start_chat(history=chat_history)
+            request_body = {
+                "contents": chat_history + [{"role": "user", "parts": [{"text":current_msg}]}],
+                "generationConfig": self.generation_config
+            }
 
-            # 发送消息并获取流式响应
+            # 构建请求URL
-            response = chat.send_message(
+            url = f"https://generativelanguage.googleapis.com/v1beta/models/{self.model_name}:generateContent?key={self.api_key}"
-                current_msg,
-                stream=True,
-                generation_config=self.generation_config
-            )
 
-            # 处理流式响应
+            # 构建请求头
-            for chunk in response:
+            headers = {
-                if hasattr(chunk, 'text') and chunk.text:
+                "Content-Type": "application/json",
-                    yield chunk.text
+            }
+
+            # 发送POST请求,经测试手动 request 无法使用 stream 模式
+            if self.proxies:
+                response = requests.post(url, headers=headers, json=request_body, stream=False, proxies=self.proxies)
+                try:
+                    data = response.json()  # 直接解析JSON
+                    if 'candidates' in data and data['candidates']:
+                        yield data['candidates'][0]['content']['parts'][0]['text']
+                    else:
+                        yield "未找到候选回复。"
+                except json.JSONDecodeError as e:
+                    yield f"JSON解码错误：{e}"
+                except Exception as e:
+                    yield f"发生错误：{e}"
+            else:
+                logger.bind(tag=TAG).info(f"Gemini stream mode ")
+                chat = self.model.start_chat(history=chat_history)
+
+                # 发送消息并获取流式响应
+                response = chat.send_message(
+                    current_msg,
+                    stream=True,
+                    generation_config=self.generation_config
+                )
+                # 处理流式响应
+                for chunk in response:
+                    if hasattr(chunk, 'text') and chunk.text:
+                        yield chunk.text
 
         except Exception as e:
             error_msg = str(e)
@@ -78,3 +124,13 @@ class LLMProvider(LLMProviderBase):
                 yield "【Gemini API key无效】"
             else:
                 yield f"【Gemini服务响应异常: {error_msg}】"
+
+
+
+
+        except requests.exceptions.RequestException as e:
+            yield f"请求失败：{e}"
+        except json.JSONDecodeError as e:
+            yield f"JSON解码错误：{e}"
+        except Exception as e:
+            yield f"发生错误：{e}"

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

@@ -1,7 +1,11 @@
 import openai
+from config.logger import setup_logging
 from core.utils.util import check_model_key
 from core.providers.llm.base import LLMProviderBase
 
+TAG = __name__
+logger = setup_logging()
+
 
 class LLMProvider(LLMProviderBase):
     def __init__(self, config):
@@ -11,6 +15,8 @@ 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)
+
         check_model_key("LLM", self.api_key)
         self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)
 
@@ -19,7 +25,8 @@ class LLMProvider(LLMProviderBase):
             responses = self.client.chat.completions.create(
                 model=self.model_name,
                 messages=dialogue,
-                stream=True
+                stream=True,
+                max_tokens=self.max_tokens,
             )
 
             is_active = True
@@ -50,7 +57,7 @@ class LLMProvider(LLMProviderBase):
                 model=self.model_name,
                 messages=dialogue,
                 stream=True,
-                tools=functions,
+                tools=functions
             )
 
             for chunk in stream: