refactor(aibox): 解耦大模型调用，新增 llm_client 抽象层以支持多厂商切换
将 think_and_respond 与 _compress_via_llm 中的直接 DeepSeek API 调用 收敛到 LLMClient 抽象层（DeepSeekClient 实现 + create_llm_client 工厂）， 统一流式事件格式，为子代理/多模型切换预留扩展点；配置新增 api.provider。
This commit is contained in:
qing committed
1 parent
417fda5479
commit
d24066369b
4 files changed
+363
-184
No files matched your search
+94
-184
@@ -35,6 +35,9 @@ from logging.handlers import RotatingFileHandler
|
||||
|
||||
import requests
|
||||
|
||||
# 大模型客户端抽象层（支持多厂商切换）
|
||||
from llm_client import create_llm_client
|
||||
|
||||
# ==================== 语言检测和多语言支持 ====================
|
||||
|
||||
|
||||
@@ -2013,7 +2016,7 @@ class ConversationCompressor:
|
||||
return content
|
||||
|
||||
def _compress_via_llm(self, messages: List[Dict]) -> Optional[str]:
|
||||
"""通过语言模型压缩消息"""
|
||||
"""通过语言模型压缩消息（经由客户端抽象层调用，支持多厂商）"""
|
||||
if not messages:
|
||||
return None
|
||||
|
||||
@@ -2022,36 +2025,15 @@ class ConversationCompressor:
|
||||
print(f"📦 {i18n.get('conversation_compress')}: 正在压缩 {len(messages)} 条消息...")
|
||||
|
||||
try:
|
||||
headers = {
|
||||
"Authorization": f"Bearer {self.ai.api_key}",
|
||||
"Content-Type": "application/json"
|
||||
}
|
||||
|
||||
payload = {
|
||||
"model": constants.DEEPSEEK_MODEL,
|
||||
"messages": [
|
||||
summary = self.ai.client.chat_completion(
|
||||
[
|
||||
{"role": "system", "content": "你是一个专业的对话摘要助手，擅长提取对话核心信息并进行结构化总结。请保留工具调用的关键结果。"},
|
||||
{"role": "user", "content": prompt}
|
||||
],
|
||||
"temperature": 0.3,
|
||||
"max_tokens": constants.COMPRESSOR_LLM_MAX_TOKENS
|
||||
}
|
||||
|
||||
response = requests.post(
|
||||
constants.DEEPSEEK_API_URL,
|
||||
headers=headers,
|
||||
json=payload,
|
||||
timeout=30
|
||||
temperature=0.3,
|
||||
max_tokens=constants.COMPRESSOR_LLM_MAX_TOKENS
|
||||
)
|
||||
|
||||
if response.status_code == 200:
|
||||
result = response.json()
|
||||
summary = result["choices"][0]["message"]["content"].strip()
|
||||
return summary
|
||||
else:
|
||||
print(f" ❌ 语言模型压缩失败: HTTP {response.status_code}")
|
||||
return None
|
||||
|
||||
return summary
|
||||
except Exception as e:
|
||||
self.logger.error(f"语言模型压缩异常: {e}")
|
||||
return None
|
||||
@@ -6580,10 +6562,13 @@ class DeepSeekChat:
|
||||
history_manager: ConversationHistory = None, memory_db: MemoryDatabase = None,
|
||||
memo_db: MemoDatabase = None, command_executor: CommandExecutor = None,
|
||||
prompt_manager: PromptManager = None):
|
||||
self.name = "DeepSeek"
|
||||
self.debug_level = debug_level if debug_level is not None else Logger("").level
|
||||
self.logger = Logger(self.name, self.debug_level)
|
||||
self.config = ConfigManager()
|
||||
self.logger = Logger("DeepSeek", debug_level if debug_level is not None else Logger("").level)
|
||||
self.debug_level = self.logger.level
|
||||
# 大模型客户端（抽象层，支持多厂商切换）
|
||||
self.client = create_llm_client(self.config, api_key=api_key, logger=self.logger)
|
||||
self.api_key = self.client.api_key
|
||||
self.name = self.client.name
|
||||
self.history_manager = history_manager
|
||||
self.memory_db = memory_db
|
||||
self.memo_db = memo_db
|
||||
@@ -6595,24 +6580,12 @@ class DeepSeekChat:
|
||||
# 添加对话压缩器
|
||||
self.compressor = ConversationCompressor(self, self.logger)
|
||||
|
||||
api_key_env = self.config.get("api.api_key_env")
|
||||
self.api_key = api_key or os.environ.get(api_key_env)
|
||||
if not self.api_key:
|
||||
try:
|
||||
home = os.environ.get("HOME", os.environ["USERPROFILE"])
|
||||
key_path = os.path.join(home, ".cert", "deepseek.key")
|
||||
with open(f"{key_path}", "r") as f:
|
||||
self.api_key = f.read().strip()
|
||||
except Exception as e:
|
||||
raise RuntimeError(f"请设置环境变量 {api_key_env} 或在配置文件中设置API Key")
|
||||
|
||||
self.max_history = max_history or constants.DEFAULT_MAX_HISTORY
|
||||
self.preserve_system = constants.PRESERVE_SYSTEM_PROMPTS
|
||||
self.stats = {"api_calls": 0, "tool_calls": 0, "errors": []}
|
||||
|
||||
# 取消机制：threading.Event 用于跨线程信号，_current_response 用于中断HTTP请求
|
||||
# 取消机制：threading.Event 用于跨线程信号，HTTP 中断由客户端抽象层负责
|
||||
self._cancel_event = threading.Event()
|
||||
self._current_response = None
|
||||
self._command_executor = command_executor # 保存命令执行器引用，用于取消时kill子进程
|
||||
|
||||
self._init_system_prompts(system_prompt)
|
||||
@@ -6624,19 +6597,10 @@ class DeepSeekChat:
|
||||
"""中断当前正在进行的AI思考/工具执行。线程安全，可从任意线程调用。"""
|
||||
self.logger.info("收到中断信号，正在取消当前操作...")
|
||||
self._cancel_event.set()
|
||||
# 强制关闭当前HTTP响应，立即中断 iter_lines() 的阻塞读取
|
||||
# 仅调用 response.close() 可能无法中断底层socket recv()，需要先关闭raw连接
|
||||
if self._current_response is not None:
|
||||
# 关闭当前流式HTTP响应，立即中断阻塞读取（由客户端抽象层负责）
|
||||
if self.client is not None:
|
||||
try:
|
||||
resp = self._current_response
|
||||
self._current_response = None
|
||||
# 先关闭底层raw socket连接，强制阻塞的recv()立即抛出异常
|
||||
if hasattr(resp, 'raw') and resp.raw is not None:
|
||||
try:
|
||||
resp.raw.close()
|
||||
except Exception:
|
||||
pass
|
||||
resp.close()
|
||||
self.client.cancel()
|
||||
except Exception:
|
||||
pass
|
||||
# 终止正在运行的子进程
|
||||
@@ -7039,11 +7003,6 @@ class DeepSeekChat:
|
||||
if self.history_manager:
|
||||
self.history_manager.add_message("user", enhanced_input, "用户")
|
||||
|
||||
headers = {
|
||||
"Authorization": f"Bearer {self.api_key}",
|
||||
"Content-Type": "application/json"
|
||||
}
|
||||
|
||||
# 构建消息列表，确保包含 reasoning_content
|
||||
enhanced_messages = []
|
||||
|
||||
@@ -7081,25 +7040,8 @@ class DeepSeekChat:
|
||||
else:
|
||||
enhanced_messages.append(msg)
|
||||
|
||||
payload = {
|
||||
"model": constants.DEEPSEEK_MODEL,
|
||||
"messages": enhanced_messages,
|
||||
"temperature": constants.DEFAULT_TEMPERATURE,
|
||||
"max_tokens": constants.DEFAULT_MAX_TOKENS,
|
||||
"stream": True
|
||||
}
|
||||
|
||||
# 思考模式：启用 thinking 参数
|
||||
if constants.REASONING_ENABLED:
|
||||
payload["thinking"] = {
|
||||
"type": "enabled"
|
||||
}
|
||||
|
||||
# 工具调用（v4-flash 支持思考+工具调用）
|
||||
tools = self.tool_registry.get_tools_schemas()
|
||||
if tools:
|
||||
payload["tools"] = tools
|
||||
payload["tool_choice"] = "auto"
|
||||
tools = self.tool_registry.get_tools_schemas() or None
|
||||
|
||||
current_time = datetime.now().strftime("%H:%M:%S")
|
||||
if output_callback and not is_continuation:
|
||||
@@ -7125,113 +7067,81 @@ class DeepSeekChat:
|
||||
MAX_CHUNK_DELAY = 0.1
|
||||
|
||||
try:
|
||||
proxies = self.config.get_proxy()
|
||||
response = requests.post(
|
||||
constants.DEEPSEEK_API_URL,
|
||||
headers=headers,
|
||||
json=payload,
|
||||
timeout=constants.DEFAULT_TIMEOUT,
|
||||
stream=True,
|
||||
proxies=proxies # 动态代理
|
||||
)
|
||||
self._current_response = response # 保存引用以便cancel()关闭连接
|
||||
|
||||
if response.status_code != 200:
|
||||
error_msg = f"\n错误状态码: {response.status_code}\n错误响应: {response.text}"
|
||||
if output_callback:
|
||||
output_callback("error", error_msg)
|
||||
response.raise_for_status()
|
||||
|
||||
for line in response.iter_lines():
|
||||
# 通过客户端抽象层发起流式请求（统一事件格式，支持多厂商切换）
|
||||
for event in self.client.stream_chat(
|
||||
enhanced_messages,
|
||||
temperature=constants.DEFAULT_TEMPERATURE,
|
||||
max_tokens=constants.DEFAULT_MAX_TOKENS,
|
||||
tools=tools,
|
||||
reasoning_enabled=constants.REASONING_ENABLED,
|
||||
):
|
||||
# 流式循环中检查取消信号
|
||||
if self._is_cancelled():
|
||||
break
|
||||
if line:
|
||||
line = line.decode('utf-8')
|
||||
if line.startswith('data: '):
|
||||
data = line[6:]
|
||||
if data == '[DONE]':
|
||||
if chunk_buffer and output_callback:
|
||||
output_callback("chunk", chunk_buffer)
|
||||
chunk_buffer = ""
|
||||
if output_callback:
|
||||
output_callback("line", "\n")
|
||||
break
|
||||
|
||||
try:
|
||||
chunk = json.loads(data)
|
||||
choices = chunk.get('choices', [])
|
||||
if not choices:
|
||||
continue
|
||||
|
||||
delta = choices[0].get('delta', {})
|
||||
|
||||
# 【关键】处理 reasoning_content（思考内容），仅当配置启用时才展示
|
||||
if constants.REASONING_ENABLED and 'reasoning_content' in delta and delta['reasoning_content']:
|
||||
reasoning = delta['reasoning_content']
|
||||
full_reasoning += reasoning
|
||||
|
||||
# 发送思考开始信号（仅第一次）
|
||||
if not reasoning_started and output_callback:
|
||||
reasoning_started = True
|
||||
|
||||
# 实时发送思考内容给前端
|
||||
if output_callback:
|
||||
output_callback("reasoning", reasoning)
|
||||
|
||||
# 思考结束：当开始收到content且之前有reasoning时，发送结束信号
|
||||
if 'content' in delta and delta['content'] and reasoning_started and not reasoning_ended:
|
||||
reasoning_ended = True
|
||||
if output_callback:
|
||||
output_callback("reasoning_end", "")
|
||||
output_callback("line", i18n.get('assistant_prefix'))
|
||||
|
||||
# 处理正常内容
|
||||
if 'content' in delta and delta['content']:
|
||||
content = delta['content']
|
||||
chunk_buffer += content
|
||||
full_response += content
|
||||
|
||||
if (len(chunk_buffer) >= MIN_CHUNK_SIZE or
|
||||
time.time() - last_send_time >= MAX_CHUNK_DELAY):
|
||||
if output_callback:
|
||||
output_callback("chunk", chunk_buffer)
|
||||
chunk_buffer = ""
|
||||
last_send_time = time.time()
|
||||
|
||||
# 处理工具调用
|
||||
if 'tool_calls' in delta:
|
||||
tool_calls = delta['tool_calls']
|
||||
for tc in tool_calls:
|
||||
index = tc.get('index', 0)
|
||||
|
||||
if index not in current_tool_calls:
|
||||
# 确保 tool_call id 非空，空 id 会导致 API 400 错误污染对话历史
|
||||
tc_id = tc.get('id', '') or f"call_{self.stats['tool_calls']}_{index}_{int(time.time()*1000)}"
|
||||
current_tool_calls[index] = {
|
||||
'id': tc_id,
|
||||
'type': 'function',
|
||||
'function': {
|
||||
'name': '',
|
||||
'arguments': ''
|
||||
}
|
||||
}
|
||||
else:
|
||||
# 流式传输中 id 可能分片到达，后续补充
|
||||
existing_id = current_tool_calls[index].get('id', '')
|
||||
new_id = tc.get('id', '')
|
||||
if new_id and (not existing_id or existing_id.startswith('call_')):
|
||||
current_tool_calls[index]['id'] = new_id
|
||||
|
||||
if 'function' in tc:
|
||||
if 'name' in tc['function']:
|
||||
current_tool_calls[index]['function']['name'] = tc['function']['name']
|
||||
if 'arguments' in tc['function']:
|
||||
current_tool_calls[index]['function']['arguments'] += tc['function']['arguments']
|
||||
|
||||
except json.JSONDecodeError as e:
|
||||
self.logger.error(f"JSON解析错误: {e}, 数据: {data}")
|
||||
continue
|
||||
|
||||
etype = event.get("type")
|
||||
|
||||
if etype == "reasoning":
|
||||
# 处理思考内容（仅当配置启用时才会收到该事件）
|
||||
reasoning = event.get("text", "")
|
||||
full_reasoning += reasoning
|
||||
if not reasoning_started and output_callback:
|
||||
reasoning_started = True
|
||||
if output_callback:
|
||||
output_callback("reasoning", reasoning)
|
||||
|
||||
elif etype == "content":
|
||||
# 思考结束：当开始收到content且之前有reasoning时，发送结束信号
|
||||
if reasoning_started and not reasoning_ended:
|
||||
reasoning_ended = True
|
||||
if output_callback:
|
||||
output_callback("reasoning_end", "")
|
||||
output_callback("line", i18n.get('assistant_prefix'))
|
||||
content = event.get("text", "")
|
||||
chunk_buffer += content
|
||||
full_response += content
|
||||
|
||||
if (len(chunk_buffer) >= MIN_CHUNK_SIZE or
|
||||
time.time() - last_send_time >= MAX_CHUNK_DELAY):
|
||||
if output_callback:
|
||||
output_callback("chunk", chunk_buffer)
|
||||
chunk_buffer = ""
|
||||
last_send_time = time.time()
|
||||
|
||||
elif etype == "tool_call":
|
||||
# 处理工具调用（流式增量）
|
||||
index = event.get("index", 0)
|
||||
if index not in current_tool_calls:
|
||||
# 确保 tool_call id 非空，空 id 会导致 API 400 错误污染对话历史
|
||||
tc_id = event.get("id") or f"call_{self.stats['tool_calls']}_{index}_{int(time.time()*1000)}"
|
||||
current_tool_calls[index] = {
|
||||
'id': tc_id,
|
||||
'type': 'function',
|
||||
'function': {
|
||||
'name': '',
|
||||
'arguments': ''
|
||||
}
|
||||
}
|
||||
else:
|
||||
# 流式传输中 id 可能分片到达，后续补充
|
||||
existing_id = current_tool_calls[index].get('id', '')
|
||||
new_id = event.get("id", "")
|
||||
if new_id and (not existing_id or existing_id.startswith('call_')):
|
||||
current_tool_calls[index]['id'] = new_id
|
||||
|
||||
if event.get("name"):
|
||||
current_tool_calls[index]['function']['name'] = event["name"]
|
||||
if event.get("arguments"):
|
||||
current_tool_calls[index]['function']['arguments'] += event["arguments"]
|
||||
|
||||
elif etype == "done":
|
||||
# 流结束：flush 剩余缓冲
|
||||
if chunk_buffer and output_callback:
|
||||
output_callback("chunk", chunk_buffer)
|
||||
chunk_buffer = ""
|
||||
if output_callback:
|
||||
output_callback("line", "\n")
|
||||
break
|
||||
|
||||
# 流式循环结束，检查是否被取消
|
||||
if self._is_cancelled():
|
||||
@@ -7373,8 +7283,8 @@ class DeepSeekChat:
|
||||
self.logger.error(f"未知错误，已回滚消息到位置 {msgs_before_turn}: {e}")
|
||||
return error_msg
|
||||
finally:
|
||||
# 清除当前响应引用
|
||||
self._current_response = None
|
||||
# 客户端抽象层在流结束时已自行清理响应引用
|
||||
pass
|
||||
|
||||
def _rollback_messages(self, to_length: int):
|
||||
"""回滚 self.messages 到指定长度，用于 API 失败时恢复对话状态。"""
|
||||
|
||||
Reference in new issue
Block a user