gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/debug'

Browse Source
This commit is contained in:
tcmofashi
2025-03-03 09:02:25 +08:00
parent f3917aea50 a7d44e3f0e
commit decd577649
26 changed files with 623 additions and 718 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
src/plugins/memory_system/llm_module.py
View File

@@ -1,66 +0,0 @@
import os
import requests
from typing import Tuple, Union
import time
from nonebot import get_driver
driver = get_driver()
config = driver.config
class LLMModel:
# def __init__(self, model_name="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", **kwargs):
def __init__(self, model_name="Pro/deepseek-ai/DeepSeek-V3", **kwargs):
self.model_name = model_name
self.params = kwargs
self.api_key = config.siliconflow_key
self.base_url = config.siliconflow_base_url
def generate_response(self, prompt: str) -> Tuple[str, str]:
"""根据输入的提示生成模型的响应"""
headers = {
"Authorization": f"Bearer {self.api_key}",
"Content-Type": "application/json"
}
# 构建请求体
data = {
"model": self.model_name,
"messages": [{"role": "user", "content": prompt}],
"temperature": 0.5,
**self.params
}
# 发送请求到完整的chat/completions端点
api_url = f"{self.base_url.rstrip('/')}/chat/completions"
max_retries = 3
base_wait_time = 15 # 基础等待时间（秒）
for retry in range(max_retries):
try:
response = requests.post(api_url, headers=headers, json=data)
if response.status_code == 429:
wait_time = base_wait_time * (2 ** retry) # 指数退避
print(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
time.sleep(wait_time)
continue
response.raise_for_status() # 检查其他响应状态
result = response.json()
if "choices" in result and len(result["choices"]) > 0:
content = result["choices"][0]["message"]["content"]
reasoning_content = result["choices"][0]["message"].get("reasoning_content", "")
return content, reasoning_content
return "没有返回结果", ""
except requests.exceptions.RequestException as e:
if retry < max_retries - 1: # 如果还有重试机会
wait_time = base_wait_time * (2 ** retry)
print(f"请求失败，等待{wait_time}秒后重试... 错误: {str(e)}")
time.sleep(wait_time)
else:
return f"请求失败: {str(e)}", ""
return "达到最大重试次数，请求仍然失败", ""

44
src/plugins/memory_system/llm_module_memory_make.py
View File

@@ -2,15 +2,17 @@ import os
import requests
from typing import Tuple, Union
import time
from ..chat.config import BotConfig
from nonebot import get_driver
import aiohttp
import asyncio
from src.plugins.chat.config import BotConfig, global_config
driver = get_driver()
config = driver.config
class LLMModel:
# def __init__(self, model_name="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", **kwargs):
def __init__(self, model_name="Pro/deepseek-ai/DeepSeek-V3", **kwargs):
def __init__(self, model_name=global_config.SILICONFLOW_MODEL_V3, **kwargs):
self.model_name = model_name
self.params = kwargs
self.api_key = config.siliconflow_key
@@ -21,7 +23,7 @@ class LLMModel:
print(f"API URL: {self.base_url}") # 打印 base_url 用于调试
def generate_response(self, prompt: str) -> Tuple[str, str]:
async def generate_response(self, prompt: str) -> Tuple[str, str]:
"""根据输入的提示生成模型的响应"""
headers = {
"Authorization": f"Bearer {self.api_key}",
@@ -44,28 +46,28 @@ class LLMModel:
for retry in range(max_retries):
try:
response = requests.post(api_url, headers=headers, json=data)
async with aiohttp.ClientSession() as session:
async with session.post(api_url, headers=headers, json=data) as response:
if response.status == 429:
wait_time = base_wait_time * (2 ** retry) # 指数退避
print(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
await asyncio.sleep(wait_time)
continue
response.raise_for_status() # 检查其他响应状态
result = await response.json()
if "choices" in result and len(result["choices"]) > 0:
content = result["choices"][0]["message"]["content"]
reasoning_content = result["choices"][0]["message"].get("reasoning_content", "")
return content, reasoning_content
return "没有返回结果", ""
if response.status_code == 429:
wait_time = base_wait_time * (2 ** retry) # 指数退避
print(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
time.sleep(wait_time)
continue
response.raise_for_status() # 检查其他响应状态
result = response.json()
if "choices" in result and len(result["choices"]) > 0:
content = result["choices"][0]["message"]["content"]
reasoning_content = result["choices"][0]["message"].get("reasoning_content", "")
return content, reasoning_content
return "没有返回结果", ""
except requests.exceptions.RequestException as e:
except Exception as e:
if retry < max_retries - 1: # 如果还有重试机会
wait_time = base_wait_time * (2 ** retry)
print(f"请求失败，等待{wait_time}秒后重试... 错误: {str(e)}")
time.sleep(wait_time)
await asyncio.sleep(wait_time)
else:
return f"请求失败: {str(e)}", ""

53
src/plugins/memory_system/memory.py
View File

@@ -1,19 +1,16 @@
# -*- coding: utf-8 -*-
import os
import jieba
from .llm_module import LLMModel
import networkx as nx
import matplotlib.pyplot as plt
import math
from collections import Counter
import datetime
import random
import time
from ..chat.config import global_config
import sys
from ...common.database import Database # 使用正确的导入语法
from ..chat.utils import calculate_information_content, get_cloest_chat_from_db
from ..models.utils_model import LLM_request
class Memory_graph:
def __init__(self):
self.G = nx.Graph() # 使用 networkx 的图结构
@@ -169,8 +166,8 @@ class Memory_graph:
class Hippocampus:
def __init__(self,memory_graph:Memory_graph):
self.memory_graph = memory_graph
self.llm_model = LLMModel()
self.llm_model_small = LLMModel(model_name="deepseek-ai/DeepSeek-V2.5")
self.llm_model = LLM_request(model = global_config.llm_normal,temperature=0.5)
self.llm_model_small = LLM_request(model = global_config.llm_normal_minor,temperature=0.5)
def get_memory_sample(self,chat_size=20,time_frequency:dict={'near':2,'mid':4,'far':3}):
current_timestamp = datetime.datetime.now().timestamp()
@@ -193,6 +190,24 @@ class Hippocampus:
chat_text.append(chat_)
return chat_text
async def memory_compress(self, input_text, rate=1):
information_content = calculate_information_content(input_text)
print(f"文本的信息量（熵）: {information_content:.4f} bits")
topic_num = max(1, min(5, int(information_content * rate / 4)))
topic_prompt = find_topic(input_text, topic_num)
topic_response = await self.llm_model.generate_response(topic_prompt)
# 检查 topic_response 是否为元组
if isinstance(topic_response, tuple):
topics = topic_response[0].split(",") # 假设第一个元素是我们需要的字符串
else:
topics = topic_response.split(",")
compressed_memory = set()
for topic in topics:
topic_what_prompt = topic_what(input_text,topic)
topic_what_response = await self.llm_model_small.generate_response(topic_what_prompt)
compressed_memory.add((topic.strip(), topic_what_response[0])) # 将话题和记忆作为元组存储
return compressed_memory
async def build_memory(self,chat_size=12):
#最近消息获取频率
time_frequency = {'near':1,'mid':2,'far':2}
@@ -208,9 +223,7 @@ class Hippocampus:
if input_text:
# 生成压缩后记忆
first_memory = set()
first_memory = self.memory_compress(input_text, 2.5)
# 延时防止访问超频
# time.sleep(5)
first_memory = await self.memory_compress(input_text, 2.5)
#将记忆加入到图谱中
for topic, memory in first_memory:
topics = segment_text(topic)
@@ -224,28 +237,6 @@ class Hippocampus:
else:
print(f"空消息 跳过")
self.memory_graph.save_graph_to_db()
def memory_compress(self, input_text, rate=1):
information_content = calculate_information_content(input_text)
print(f"文本的信息量（熵）: {information_content:.4f} bits")
topic_num = max(1, min(5, int(information_content * rate / 4)))
# print(topic_num)
topic_prompt = find_topic(input_text, topic_num)
topic_response = self.llm_model.generate_response(topic_prompt)
# 检查 topic_response 是否为元组
if isinstance(topic_response, tuple):
topics = topic_response[0].split(",") # 假设第一个元素是我们需要的字符串
else:
topics = topic_response.split(",")
# print(topics)
compressed_memory = set()
for topic in topics:
if topic=='' or '[' in topic:
continue
topic_what_prompt = topic_what(input_text,topic)
topic_what_response = self.llm_model_small.generate_response(topic_what_prompt)
compressed_memory.add((topic.strip(), topic_what_response[0])) # 将话题和记忆作为元组存储
return compressed_memory
def segment_text(text):