feat: 优化 FAQ 处理和系统稳定性
- 添加本地 FAQ 库快速路径（问候语等社交响应） - 修复 Chatwoot 重启循环问题（PID 文件清理） - 添加 LLM 响应缓存（Redis 缓存，提升性能） - 添加智能推理模式（根据查询复杂度自动启用） - 添加订单卡片消息功能（Chatwoot 富媒体） - 增加 LLM 超时时间至 60 秒 Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit is contained in:
wangliang
@@ -9,6 +9,7 @@ from zhipuai import ZhipuAI
|
||||
|
||||
from config import settings
|
||||
from utils.logger import get_logger
|
||||
from utils.response_cache import get_response_cache
|
||||
|
||||
logger = get_logger(__name__)
|
||||
|
||||
@@ -31,19 +32,80 @@ class LLMResponse:
|
||||
class ZhipuLLMClient:
|
||||
"""ZhipuAI LLM Client wrapper"""
|
||||
|
||||
DEFAULT_TIMEOUT = 30 # seconds
|
||||
DEFAULT_TIMEOUT = 60 # seconds (increased from 30 for better reliability)
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
api_key: Optional[str] = None,
|
||||
model: Optional[str] = None,
|
||||
timeout: Optional[int] = None
|
||||
timeout: Optional[int] = None,
|
||||
enable_reasoning: Optional[bool] = None
|
||||
):
|
||||
self.api_key = api_key or settings.zhipu_api_key
|
||||
self.model = model or settings.zhipu_model
|
||||
self.timeout = timeout or self.DEFAULT_TIMEOUT
|
||||
self.enable_reasoning = enable_reasoning if enable_reasoning is not None else settings.enable_reasoning_mode
|
||||
self._client = ZhipuAI(api_key=self.api_key)
|
||||
logger.info("ZhipuAI client initialized", model=self.model, timeout=self.timeout)
|
||||
logger.info(
|
||||
"ZhipuAI client initialized",
|
||||
model=self.model,
|
||||
timeout=self.timeout,
|
||||
reasoning_mode=self.enable_reasoning
|
||||
)
|
||||
|
||||
def _should_use_reasoning(self, messages: list[dict[str, str]]) -> bool:
|
||||
"""Determine if reasoning mode should be used based on query complexity
|
||||
|
||||
Args:
|
||||
messages: List of message dictionaries
|
||||
|
||||
Returns:
|
||||
True if reasoning mode should be used
|
||||
"""
|
||||
if not self.enable_reasoning:
|
||||
return False
|
||||
|
||||
if not settings.reasoning_mode_for_complex:
|
||||
# If smart mode is disabled, use the global setting
|
||||
return self.enable_reasoning
|
||||
|
||||
# Smart mode: analyze the last user message
|
||||
last_message = ""
|
||||
for msg in reversed(messages):
|
||||
if msg.get("role") == "user":
|
||||
last_message = msg.get("content", "")
|
||||
break
|
||||
|
||||
# Simple queries that don't need reasoning
|
||||
simple_patterns = [
|
||||
"你好", "hi", "hello", "嗨",
|
||||
"谢谢", "thank", "感谢",
|
||||
"再见", "bye", "拜拜",
|
||||
"退货政策", "营业时间", "联系方式",
|
||||
"发货", "配送", "物流"
|
||||
]
|
||||
|
||||
last_message_lower = last_message.lower()
|
||||
for pattern in simple_patterns:
|
||||
if pattern in last_message_lower:
|
||||
logger.debug("Simple query detected, disabling reasoning", query=last_message[:50])
|
||||
return False
|
||||
|
||||
# Complex queries that benefit from reasoning
|
||||
complex_patterns = [
|
||||
"为什么", "how", "why", "如何",
|
||||
"推荐", "recommend", "建议",
|
||||
"比较", "compare", "区别",
|
||||
"怎么样", "如何选择"
|
||||
]
|
||||
|
||||
for pattern in complex_patterns:
|
||||
if pattern in last_message_lower:
|
||||
logger.debug("Complex query detected, enabling reasoning", query=last_message[:50])
|
||||
return True
|
||||
|
||||
# Default: disable reasoning for speed
|
||||
return False
|
||||
|
||||
async def chat(
|
||||
self,
|
||||
@@ -51,14 +113,39 @@ class ZhipuLLMClient:
|
||||
temperature: float = 0.7,
|
||||
max_tokens: int = 2048,
|
||||
top_p: float = 0.9,
|
||||
use_cache: bool = True,
|
||||
enable_reasoning: Optional[bool] = None,
|
||||
**kwargs: Any
|
||||
) -> LLMResponse:
|
||||
"""Send chat completion request"""
|
||||
"""Send chat completion request with caching support"""
|
||||
formatted_messages = [
|
||||
{"role": msg.role, "content": msg.content}
|
||||
for msg in messages
|
||||
]
|
||||
|
||||
# Try cache first
|
||||
if use_cache:
|
||||
try:
|
||||
cache = get_response_cache()
|
||||
cached_response = await cache.get(
|
||||
model=self.model,
|
||||
messages=formatted_messages,
|
||||
temperature=temperature
|
||||
)
|
||||
if cached_response is not None:
|
||||
logger.info(
|
||||
"Returning cached response",
|
||||
model=self.model,
|
||||
response_length=len(cached_response)
|
||||
)
|
||||
return LLMResponse(
|
||||
content=cached_response,
|
||||
finish_reason="cache_hit",
|
||||
usage={"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
|
||||
)
|
||||
except Exception as e:
|
||||
logger.warning("Cache check failed", error=str(e))
|
||||
|
||||
logger.info(
|
||||
"Sending chat request",
|
||||
model=self.model,
|
||||
@@ -66,15 +153,32 @@ class ZhipuLLMClient:
|
||||
temperature=temperature
|
||||
)
|
||||
|
||||
# Determine if reasoning mode should be used
|
||||
use_reasoning = enable_reasoning if enable_reasoning is not None else self._should_use_reasoning(formatted_messages)
|
||||
|
||||
if use_reasoning:
|
||||
logger.info("Reasoning mode enabled for this request")
|
||||
|
||||
def _make_request():
|
||||
return self._client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=formatted_messages,
|
||||
temperature=temperature,
|
||||
max_tokens=max_tokens,
|
||||
top_p=top_p,
|
||||
**kwargs
|
||||
)
|
||||
request_params = {
|
||||
"model": self.model,
|
||||
"messages": formatted_messages,
|
||||
"temperature": temperature,
|
||||
"max_tokens": max_tokens,
|
||||
"top_p": top_p,
|
||||
}
|
||||
|
||||
# Add thinking mode control
|
||||
# Format: {"thinking": {"type": "disabled"}} or {"type": "enabled"}
|
||||
if use_reasoning:
|
||||
request_params["thinking"] = {"type": "enabled"}
|
||||
logger.info("Thinking mode: enabled", request_params={"thinking": {"type": "enabled"}})
|
||||
else:
|
||||
request_params["thinking"] = {"type": "disabled"}
|
||||
logger.info("Thinking mode: disabled", request_params={"thinking": {"type": "disabled"}})
|
||||
|
||||
request_params.update(kwargs)
|
||||
return self._client.chat.completions.create(**request_params)
|
||||
|
||||
try:
|
||||
with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
|
||||
@@ -94,6 +198,19 @@ class ZhipuLLMClient:
|
||||
if not content:
|
||||
logger.warning("LLM returned empty content")
|
||||
|
||||
# Cache the response
|
||||
if use_cache and content:
|
||||
try:
|
||||
cache = get_response_cache()
|
||||
await cache.set(
|
||||
model=self.model,
|
||||
messages=formatted_messages,
|
||||
response=content,
|
||||
temperature=temperature
|
||||
)
|
||||
except Exception as e:
|
||||
logger.warning("Failed to cache response", error=str(e))
|
||||
|
||||
return LLMResponse(
|
||||
content=content or "",
|
||||
finish_reason=choice.finish_reason,
|
||||
|
||||
Reference in New Issue
Block a user