cyonjan/PyWxDump
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
6bb8b8b6ec
PyWxDump/pywxdump/api/api_utils/llm.py

477 lines
11 KiB
Python
Raw Normal View History

完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
# LLM api相关
import enum
import json
import os
import re
import httpx
ai可视化后端完成
2025-05-02 20:38:57 +08:00
import openai
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
from openai import OpenAI
from pywxdump.api.remote_server import gc
class BaseLLMApi(object):
def __init__(self,api_key,base_url=None):
# 设置名字，以供其他函数使用 !!!不使用，
# self.api_key_string = "API_KEY"
# self.base_url_string = "BASE_URL"
# self.env_api_key_string = self.__class__.__name__ + "_" + self.api_key_string
# self.env_base_url_string = self.__class__.__name__ + "_" + self.base_url_string
# self.setting_string = self.__class__.__name__ + "_setting"
self.API_KEY = api_key
self.BASE_URL = base_url
self.module = (
)#模型列表
self.HTTP_CLIENT = None
self.isReady = False
self.message = []
# 执行初始化方法
self.set_default_fn()
def set_default_fn(self):
if not self.module:
self.set_default_module()
if not self.BASE_URL:
self.set_default_base_url()
if not self.message:
self.set_default_message()
def set_default_module(self):
self.module = ()
def set_default_base_url(self):
self.BASE_URL = ""
def set_default_message(self):
ai可视化后端完成
2025-05-02 20:38:57 +08:00
"""
要确保message中至少有两个元素，第一个元素为系统消息，第二个元素为用户消息，且第二个元素中有{{content}}
"""
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
self.message = [
{"role": "system", "content": "You are a helpful assistant"},
ai可视化后端完成
2025-05-02 20:38:57 +08:00
{"role": "user", "content": "Hello {{content}}"},
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
]
def set_module(self, module):
self.module = module
def set_api_key(self, api_key):
self.API_KEY = api_key
def set_base_url(self, base_url):
self.BASE_URL = base_url
def set_message(self, message):
self.message = message
# def ready(self):
# if not self.API_KEY:
# # 从配置中获取，这个功能必须配合网页API开启后才能使用
# self.API_KEY = gc.get_conf(gc.at,self.setting_string)[self.api_key_string]
# if not self.API_KEY:
# raise RuntimeError("API_KEY must be set")
# # 设置环境变量
# os.environ[self.env_api_key_string] = self.API_KEY
# os.environ[self.env_base_url_string] = self.BASE_URL
#
# self.isReady = True
# return
def ready(self):
if not self.BASE_URL and not self.API_KEY:
raise RuntimeError("API_KEY or BASE_URL must be set")
self.isReady = True
@property
def http_client(self):
if not self.isReady:
self.ready()
try:
self.HTTP_CLIENT = OpenAI(api_key=self.API_KEY, base_url=self.BASE_URL)
return self.HTTP_CLIENT
except:
raise RuntimeError("HTTP_CLIENT set not successfully,please check!")
def send_msg(self, message=None, module=None, stream=False):
"""
向大模型发送信息
如果非流式返回，则直接输出内容，
否则使用openai文档规定格式输出
"""
if message is None:
ai可视化后端完成
2025-05-02 20:38:57 +08:00
self.message[1]["content"] = self.message[1].get("content").replace("{{content}}", " ")
message = self.message
else:
self.message[1]["content"] = self.message[1].get("content").replace("{{content}}", message)
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
message = self.message
response = self.http_client.chat.completions.create(
model=self.module[module],
messages=message,
stream=stream
)
if not stream:
return self.process_msg(response.choices[0].message.content)
else:
return self.process_msg(response.response.read().decode("utf-8"))
ai可视化后端完成
2025-05-02 20:38:57 +08:00
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
def process_msg(self,x):
return x
class DeepSeekApi(BaseLLMApi):
def set_default_module(self):
self.module = (
"deepseek-chat",
"deepseek-reasoner"
)
def set_default_base_url(self):
self.BASE_URL = "https://api.deepseek.com"
def set_default_message(self):
self.message = [
{"role": "system", "content": """从内容中提取出以下信息，可以根据内容多少进行列表扩展或增加，请仔细思索怎么填充内容，如果没有给到合理的名称或其他内容，就以合理的方式思考并添加。
- 最后的输出值使用严格的json格式
- 不要私自添加json块或减少json块
- 内容中不要使用换行符，如果内容原本有多个换行符，删掉原本多余的的换行符，只保留一个换行符再加入进去。
- 内容中如果有很奇怪的字符，比如''\''或''\\''影响代码编译的字符，删除原本的字符再加入进去。
{
"header": {
"title": "[群名称]报告",
"date": "[日期]",
"metaInfo": {
"totalMessages": "[数量]",
"activeUsers": "[数量]",
"timeRange": "[时间范围]"
}
},
"sections": {
"hotTopics": {
"items": [
{
"name": "[热点话题名称]",
"category": "[话题分类]",
"summary": "[简要总结(50-100字)]",
"keywords": ["[关键词1]", "[关键词2]"],
"mentions": "[次数]"
}
]
},
"tutorials": {
"items": [
{
"type": "[TUTORIAL | NEWS | RESOURCE]",
"title": "[分享的教程或资源标题]",
"sharedBy": "[昵称]",
"time": "[时间]",
"summary": "[内容简介]",
"keyPoints": ["[要点1]", "[要点2]"],
"url": "[URL]",
"domain": "[域名]",
"category": "[分类]"
}
]
},
"importantMessages": {
"items": [
{
"time": "[消息时间]",
"sender": "[发送者昵称]",
"type": "[NOTICE | EVENT | ANNOUNCEMENT | OTHER]",
"priority": "[高|中|低]",
"content": "[消息内容]",
"fullContent": "[完整通知内容]"
}
]
},
"dialogues": {
"items": [
{
"type": "[DIALOGUE | QUOTE]",
"messages": [
{
"speaker": "[说话者昵称]",
"time": "[发言时间]",
"content": "[消息内容]"
}
],
"highlight": "[对话中的金句或亮点]",
"relatedTopic": "[某某话题]"
}
]
},
"qa": {
"items": [
{
"question": {
"asker": "[提问者昵称]",
"time": "[提问时间]",
"content": "[问题内容]",
"tags": ["[相关标签1]", "[相关标签2]"]
},
"answers": [
{
"responder": "[回答者昵称]",
"time": "[回答时间]",
"content": "[回答内容]",
"isAccepted": true
}
]
}
]
},
"analytics": {
"heatmap": [
{
"topic": "[话题名称]",
"percentage": "[百分比]",
"color": "#3da9fc",
"count": "[数量]"
}
],
"chattyRanking": [
{
"rank": 1,
"name": "[群友昵称]",
"count": "[数量]",
"characteristics": ["[特点1]", "[特点2]"],
"commonWords": ["[常用词1]", "[常用词2]"]
}
],
"nightOwl": {
"name": "[熬夜冠军昵称]",
"title": "[熬夜冠军称号]",
"latestTime": "[时间]",
"messageCount": "[数量]",
"lastMessage": "[最后一条深夜消息内容]"
}
},
"wordCloud": {
"words": [
{
"text": "[关键词1]",
"size": 38,
"color": "#00b4d8",
"rotation": -15
}
],
"legend": [
{"color": "#00b4d8", "label": "[分类1] 相关词汇"},
{"color": "#4361ee", "label": "[分类2] 相关词汇"}
]
}
},
"footer": {
"dataSource": "[群名称]聊天记录",
"generationTime": "[当前时间]",
"statisticalPeriod": "[日期] [时间范围]",
"disclaimer": "本报告内容基于群聊公开讨论，如有不当内容或侵权问题请联系管理员处理。"
}
}"""},
ai可视化后端完成
2025-05-02 20:38:57 +08:00
{"role": "user", "content": """你好，以下是我要提取的内容: {{content}}"""},
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
]
def process_msg(self,x):
"""
识别json格式，并返回字典
"""
pattern = re.compile('{.*}', flags=re.IGNORECASE | re.MULTILINE | re.S)
# print(pattern.search(json_data).group())
json_data = json.loads(pattern.search(x).group())
return json_data
if __name__ == "__main__":
deepseek_api = DeepSeekApi("sk-2ed4377a895d4ce18e086258c254fc8e")
ai可视化后端完成
2025-05-02 20:38:57 +08:00
response = deepseek_api.send_msg(module=0,message="""""")
完成了LLM的适配，DIfy暂时不用
2025-05-01 18:14:28 +08:00
print(response)
Reference in New Issue Copy Permalink