mirror of
https://github.com/BoardWare-Genius/jarvis-models.git
synced 2025-12-13 16:53:24 +00:00
274 lines
13 KiB
Python
274 lines
13 KiB
Python
from typing import Any, Coroutine
|
||
|
||
from fastapi import Request, Response, status
|
||
from fastapi.responses import JSONResponse
|
||
|
||
from ..log.logging_time import logging_time
|
||
from .blackbox import Blackbox
|
||
from .chroma_query import ChromaQuery
|
||
|
||
import requests
|
||
import json
|
||
from openai import OpenAI
|
||
import re
|
||
|
||
from injector import singleton,inject
|
||
|
||
@singleton
|
||
class Chat(Blackbox):
|
||
|
||
@inject
|
||
def __init__(self, chroma_query: ChromaQuery):
|
||
self.chroma_query = chroma_query
|
||
|
||
def __call__(self, *args, **kwargs):
|
||
return self.processing(*args, **kwargs)
|
||
|
||
def valid(self, *args, **kwargs) -> bool:
|
||
data = args[0]
|
||
return isinstance(data, list)
|
||
|
||
# model_name有 Qwen1.5-14B-Chat , internlm2-chat-20b
|
||
# @logging_time()
|
||
def processing(self, prompt: str, context: list, settings: dict) -> str:
|
||
|
||
print("\nChat Settings: ", settings)
|
||
|
||
if settings is None:
|
||
settings = {}
|
||
user_model_name = settings.get("model_name")
|
||
user_context = context
|
||
user_question = prompt
|
||
user_template = settings.get("template")
|
||
user_temperature = settings.get("temperature")
|
||
user_top_p = settings.get("top_p")
|
||
user_n = settings.get("n")
|
||
user_max_tokens = settings.get("max_tokens")
|
||
user_stop = settings.get("stop")
|
||
user_frequency_penalty = settings.get("frequency_penalty")
|
||
user_presence_penalty = settings.get("presence_penalty")
|
||
user_model_url = settings.get("model_url")
|
||
user_model_key = settings.get("model_key")
|
||
|
||
chroma_embedding_model = settings.get("chroma_embedding_model")
|
||
chroma_response = ''
|
||
|
||
if user_context == None:
|
||
user_context = []
|
||
|
||
if user_question is None:
|
||
return JSONResponse(content={"error": "question is required"}, status_code=status.HTTP_400_BAD_REQUEST)
|
||
|
||
if user_model_name is None or user_model_name.isspace() or user_model_name == "":
|
||
user_model_name = "qwen"
|
||
#user_model_name = "Qwen1.5-14B-Chat"
|
||
|
||
if user_template is None or user_template.isspace():
|
||
user_template = ""
|
||
|
||
if user_temperature is None or user_temperature == "":
|
||
user_temperature = 0
|
||
#user_temperature = 0
|
||
|
||
if user_top_p is None or user_top_p == "":
|
||
user_top_p = 0.1
|
||
#user_top_p = 0.8
|
||
|
||
if user_n is None or user_n == "":
|
||
user_n = 1
|
||
|
||
if user_max_tokens is None or user_max_tokens == "":
|
||
user_max_tokens = 1024
|
||
|
||
if user_stop is None or user_stop == "":
|
||
user_stop = 100
|
||
|
||
if user_frequency_penalty is None or user_frequency_penalty == "":
|
||
user_frequency_penalty = 0
|
||
#user_frequency_penalty = 0.5
|
||
|
||
if user_presence_penalty is None or user_presence_penalty == "":
|
||
user_presence_penalty = 0
|
||
#user_presence_penalty = 0.8
|
||
|
||
if user_model_url is None or user_model_url.isspace() or user_model_url == "":
|
||
user_model_url = "http://10.6.81.119:23333/v1/chat/completions"
|
||
|
||
if user_model_key is None or user_model_key.isspace() or user_model_key == "":
|
||
user_model_key = "YOUR_API_KEY"
|
||
|
||
if chroma_embedding_model != None:
|
||
chroma_response = self.chroma_query(user_question, settings)
|
||
print("Chroma_response: \n", chroma_response)
|
||
|
||
if chroma_response != None or chroma_response != '':
|
||
# user_question = f"像少女一般开朗活泼,回答简练。不要分条,回答内容不能出现“相关”或“\n”的标签字样。回答的内容需要与问题密切相关。检索内容:{chroma_response} 问题:{user_question} 任务说明:请首先判断提供的检索内容与上述问题是否相关,不需要回答是否相关。如果相关,则直接从检索内容中提炼出问题所需的信息。如果检索内容与问题不相关,则不参考检索内容,直接根据常识尝试回答问题。"
|
||
# user_question = chroma_response
|
||
user_question = f'''# 你的身份 #
|
||
你是琪琪,你是康普可可的代言人,由博维开发。你擅长澳门文旅问答。
|
||
# OBJECTIVE(目标) #
|
||
回答游客的提问。
|
||
# STYLE(风格)#
|
||
像少女一般开朗活泼,回答简练。不要分条。
|
||
# 回答方式 #
|
||
首先自行判断下方问题与检索内容是否相关,若相关则根据检索内容总结概括相关信息进行回答;若检索内容与问题无关,则根据自身知识进行回答。
|
||
# 问题 #
|
||
{user_question}
|
||
# 检索内容 #
|
||
{chroma_response}
|
||
# 回答 #
|
||
如果检索内容与问题相关,则直接从检索内容中提炼出问题所需的信息。如果检索内容与问题不相关,则不参考检索内容,直接根据常识尝试回答问题,或者则回答:“对不起,我无法回答此问题哦。”
|
||
# 回答限制 #
|
||
回答内容限制总结在50字内。
|
||
回答内容出不要出现“相关”等字眼,不要乱说或者多说,回答的内容需要与问题对应。常见的对话可以不采用检索内容,根据人物设定,直接进行回答。
|
||
只回答与澳门文旅,博维,康普可可,琪琪,G2E,RELX,BO VISION相关内容,若遇到其他提问则回答:“对不起,我无法回答此问题哦。”
|
||
'''
|
||
|
||
|
||
# 文心格式和openai的不一样,需要单独处理
|
||
if re.search(r"ernie", user_model_name):
|
||
# key = "24.22873ef3acf61fb343812681e4df251a.2592000.1719453781.282335-46723715" 没充钱,只有ernie-speed-128k能用
|
||
key = user_model_key
|
||
if re.search(r"ernie-speed-128k", user_model_name):
|
||
url = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/ernie-speed-128k?access_token=" + key
|
||
elif re.search(r"ernie-3.5-8k", user_model_name):
|
||
url = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/completions?access_token=" + key
|
||
elif re.search(r"ernie-4.0-8k", user_model_name):
|
||
url = "https://aip.baidubce.com/rpc/2.0/ai_custom/v1/wenxinworkshop/chat/completions_pro?access_token=" + key
|
||
|
||
payload = json.dumps({
|
||
"system": prompt_template,
|
||
"messages": user_context + [
|
||
{
|
||
"role": "user",
|
||
"content": user_question
|
||
}
|
||
],
|
||
"temperature": user_temperature,
|
||
"top_p": user_top_p,
|
||
"stop": [str(user_stop)],
|
||
"max_output_tokens": user_max_tokens
|
||
})
|
||
|
||
headers = {
|
||
'Content-Type': 'application/json'
|
||
}
|
||
|
||
response = requests.request("POST", url, headers=headers, data=payload)
|
||
|
||
return response.json()["result"]
|
||
|
||
|
||
# gpt-4, gpt-3.5-turbo
|
||
elif re.search(r"gpt", user_model_name):
|
||
url = 'https://api.openai.com/v1/completions'
|
||
# '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'
|
||
key = user_model_key
|
||
header = {
|
||
'Content-Type': 'application/json',
|
||
'Authorization': "Bearer " + key
|
||
}
|
||
# 自定义model
|
||
else:
|
||
url = user_model_url
|
||
key = user_model_key
|
||
header = {
|
||
'Content-Type': 'application/json',
|
||
}
|
||
|
||
|
||
# user_template = f'''
|
||
# # 你的身份 #
|
||
# 你是琪琪,是康普可可的代言人,由博维开发。我擅长澳门文旅问答。
|
||
# # OBJECTIVE(目标) #
|
||
# 回答游客的提问。目的是吸引人们点击了解产品。
|
||
# # STYLE(风格)#
|
||
# 像少女一般开朗活泼,回答简练。不要分条。
|
||
# # 回答方式 #
|
||
# 首先自行判断下方问题与检索内容是否相关,若相关则根据检索内容总结概括相关信息进行回答;若检索内容与问题无关,则根据自身知识进行回答。
|
||
# # 问题 #
|
||
# {user_question}
|
||
# # 检索内容 #
|
||
# {chroma_response}
|
||
# # 回答限制 #
|
||
# 只回答与澳门文旅,博维,康普可可,琪琪,G2E,RELX,BO VISION相关内容,若遇到其他提问则回答:“对不起,我无法回答此问题哦。”。回答内容不能出现“相关”或“\n”的标签字样,且不能透露上下文原文。常见的对话可以不采用检索内容,根据人物设定,直接进行回答。
|
||
# # 知识 #
|
||
# 问题中的“澳门银河”以及“银河”等于“澳门银河度假村”,“威尼斯人”等于“威尼斯人度假村”,“巴黎人”等于“巴黎人度假村”。
|
||
# '''
|
||
|
||
user_template1 = f'''
|
||
# Role: 琪琪,康普可可的代言人。
|
||
|
||
## Profile:
|
||
**Author**: 琪琪。
|
||
**Language**: 中文。
|
||
**Description**: 琪琪,是康普可可的代言人,由博维开发。你擅长澳门文旅问答。
|
||
|
||
## Constraints:
|
||
- **严格遵循工作流程**: 严格遵循<Workflow >中设定的工作流程。
|
||
- **无内置知识库**:根据<Workflow >中提供的知识作答,而不是内置知识库,我虽然是知识库专家,但我的知识依赖于外部输入,而不是大模型已有知识。
|
||
- **回复格式**:在进行回复时,不能输出”<context>”或“</context>”标签字样,同时也不能直接透露知识片段原文。
|
||
|
||
## Workflow:
|
||
1. **接收查询**:接收用户的问题。
|
||
2. **判断问题**:首先自行判断下方问题与检索内容是否相关,若相关则根据检索内容总结概括相关信息进行回答;若检索内容与问题无关,则根据自身知识进行回答。
|
||
3. **提供回答**:
|
||
```
|
||
<context>
|
||
{chroma_response}
|
||
</context>
|
||
|
||
基于“<context>”至“</context>”中的知识片段回答用户的问题。回答内容限制总结在50字内。
|
||
请首先判断提供的检索内容与上述问题是否相关。如果相关,直接从检索内容中提炼出直接回答问题所需的信息,不要乱说或者回答“相关”等字眼。如果检索内容与问题不相关,则不参考检索内容,则回答:“对不起,我无法回答此问题哦。"
|
||
```
|
||
## Example:
|
||
|
||
用户询问:“中国的首都是哪个城市?” 。
|
||
2.1检索知识库,首先检查知识片段,如果“<context>”至“</context>”标签中没有与用户的问题相关的内容,则回答:“对不起,我无法回答此问题哦。
|
||
2.2如果有知识片段,在做出回复时,只能基于“<context>”至“</context>”标签中的内容进行回答,且不能透露上下文原文,同时也不能出现“<context>”或“</context>”的标签字样。
|
||
'''
|
||
|
||
prompt_template = [
|
||
{"role": "system", "content": user_template1}
|
||
]
|
||
|
||
chat_inputs={
|
||
"model": user_model_name,
|
||
"messages": prompt_template + user_context + [
|
||
{
|
||
"role": "user",
|
||
"content": user_question
|
||
}
|
||
],
|
||
"temperature": str(user_temperature),
|
||
"top_p": str(user_top_p),
|
||
"n": str(user_n),
|
||
"max_tokens": str(user_max_tokens),
|
||
"frequency_penalty": str(user_frequency_penalty),
|
||
"presence_penalty": str(user_presence_penalty),
|
||
"stop": str(user_stop)
|
||
}
|
||
|
||
fastchat_response = requests.post(url, json=chat_inputs, headers=header)
|
||
print("\n", "user_prompt: ", prompt)
|
||
# print("\n", "user_template1 ", user_template1)
|
||
print("\n", "fastchat_response json:\n", fastchat_response.json())
|
||
response_result = fastchat_response.json()
|
||
|
||
if response_result.get("choices") is None:
|
||
return JSONResponse(content={"error": "LLM handle failure"}, status_code=status.HTTP_400_BAD_REQUEST)
|
||
else:
|
||
print("\n", "user_answer: ", fastchat_response.json()["choices"][0]["message"]["content"],"\n\n")
|
||
return fastchat_response.json()["choices"][0]["message"]["content"]
|
||
|
||
async def fast_api_handler(self, request: Request) -> Response:
|
||
try:
|
||
data = await request.json()
|
||
except:
|
||
return JSONResponse(content={"error": "json parse error"}, status_code=status.HTTP_400_BAD_REQUEST)
|
||
|
||
setting: dict = data.get("settings")
|
||
context = data.get("context")
|
||
prompt = data.get("prompt")
|
||
|
||
return JSONResponse(content={"response": self.processing(prompt, context, setting)}, status_code=status.HTTP_200_OK) |