本文主要是介绍RAG - QA + Qwen + dashscope,希望对大家解决编程问题提供一定的参考价值,需要的开发者们随着小编来一起学习吧!
文章目录
- 一、关于项目
- 二、准备
- 1、安装依赖包
- 2、准备数据集
- 3、准备 API-Key
- 三、代码实现
- 准备数据
- 生成 embeddings
- 检索方法
- 问答
转载改编自:qwen_doc_search_QA_based_on_dashscope.ipynb
https://github.com/modelscope/modelscope/blob/master/examples/pytorch/application/qwen_doc_search_QA_based_on_dashscope.ipynb
一、关于项目
- dashscope : https://blog.csdn.net/lovechris00/article/details/136938486
- Qwen : https://blog.csdn.net/lovechris00/article/details/135790058
二、准备
1、安装依赖包
# install required packages
!pip install dashvector dashscope
!pip install transformers_stream_generator python-dotenv
2、准备数据集
这里使用的是 中文突发事件语料库,由 上海大学-语义智能实验室 提供
https://github.com/shijiebei2009/CEC-Corpus
# prepare news corpus as knowledge source
!git clone https://github.com/shijiebei2009/CEC-Corpus.git
数据集内容:
../datasets/CEC-Corpus$ tree
.
├── CEC
│ ├── 交通事故
│ │ ├── 101国道密云段现惨祸客车农用车相撞致6人亡.xml
│ │ ├── 104国道浙江温岭段发生翻车事故致2死2伤.xml
│ │ ├── ...
│ │ └── 黑龙江五常发生特大交通事故6人死亡.xml
│ ├── 地震
│ │ ├── 上海:高层建筑普遍有震感但不会造成危害.xml
│ │ ├── ...
│ │ ├── 重庆市区有明显震感电线杆在摇晃.xml
│ │ └── 青海发生6.3级地震震区人口密度低尚无人员伤亡.xml
│ ├── 恐怖袭击
│ │ ├── 4月7日凌晨5时,近300名穿着“警察”制服.xml
│ │ ├── 世界杯险遇恐怖袭击警方发现犯罪组织欲炸桥.xml
│ │ ├── ...
│ │ └── 阿尔及利亚汽车炸弹爆炸11人死31人伤.xml
│ ├── 火灾
│ │ ├── 上海永嘉路老式洋房突发火灾好心市民合力救出被困老太.xml
│ │ ├── 云南丽江束河古镇昨凌晨失火.xml
│ │ ├── ...
│ │ └── 马尼拉华人区住宅发生火灾一华人老妇被烧伤.xml
│ └── 食物中毒
│ ├── 上海一家公司70多名员工食物中毒.xml
│ ├── ...
│ └── 龙岗一小食店发生一起疑似中毒事件.xml
├── raw corpus (332 文件)
│ └── allSourceText
│ ├── 101国道密云段现惨祸客车农用车相撞致6人亡.txt
│ ├── ...
│ ├── 黑龙江鸡西市20多名小学生疑似食物中毒.txt
│ └── 龙岗一小食店发生一起疑似中毒事件.txt
└── README.md
3、准备 API-Key
如果没有,可以点击申请;
阿里云需要实名认证后才能刚申请,人脸识别也很快。
- dashvector API-Key
https://dashvector.console.aliyun.com/cn-hangzhou/api-key - dashscope API-Key
https://dashscope.console.aliyun.com/apiKey
三、代码实现
import dashscope
import os
from dotenv import load_dotenv
from dashscope import TextEmbedding
from dashvector import Client, Doc# get env variable from .env
# please make sure DASHSCOPE_KEY is defined in .env
load_dotenv()
dashscope.api_key = os.getenv('DASHSCOPE_KEY')# initialize DashVector for embedding's indexing and searching
dashvector_client = Client(api_key='{your-dashvector-api-key}')# define collection name
collection_name = 'news_embeddings'# delete if already exist
dashvector_client.delete(collection_name)# create a collection with embedding size of 1536
rsp = dashvector_client.create(collection_name, 1536)
collection = dashvector_client.get(collection_name)
准备数据
def prepare_data_from_dir(path, size):# prepare the data from a file folder in order to upsert to DashVector with a reasonable doc's size.batch_docs = []for file in os.listdir(path):with open(path + '/' + file, 'r', encoding='utf-8') as f:batch_docs.append(f.read())if len(batch_docs) == size:yield batch_docs[:]batch_docs.clear()if batch_docs:yield batch_docs
def prepare_data_from_file(path, size):# prepare the data from file in order to upsert to DashVector with a reasonable doc's size.batch_docs = []chunk_size = 12with open(path, 'r', encoding='utf-8') as f:doc = ''count = 0for line in f:if count < chunk_size and line.strip() != '':doc += linecount += 1if count == chunk_size:batch_docs.append(doc)if len(batch_docs) == size:yield batch_docs[:]batch_docs.clear()doc = ''count = 0if batch_docs:yield batch_docs
生成 embeddings
def generate_embeddings(docs):# create embeddings via DashScope's TextEmbedding model APIrsp = TextEmbedding.call(model=TextEmbedding.Models.text_embedding_v1,input=docs)embeddings = [record['embedding'] for record in rsp.output['embeddings']]return embeddings if isinstance(docs, list) else embeddings[0]
id = 0
dir_path = 'xx/CEC-Corpus/raw corpus/allSourceText'# indexing the raw docs with index to DashVector
collection = dashvector_client.get(collection_name)# embedding api max batch size
batch_size = 4 for news in list(prepare_data_from_dir(dir_path, batch_size)):ids = [id + i for i, _ in enumerate(news)]id += len(news)# generate embedding from raw docsvectors = generate_embeddings(news)# upsert and indexret = collection.upsert([Doc(id=str(id), vector=vector, fields={"raw": doc})for id, doc, vector in zip(ids, news, vectors)])print(ret)
# check the collection status
collection = dashvector_client.get(collection_name)
rsp = collection.stats()
print(rsp)
检索方法
def search_relevant_context(question, topk=1, client=dashvector_client):# query and recall the relevant informationcollection = client.get(collection_name)# recall the top k similarity results from DashVectorrsp = collection.query(generate_embeddings(question), output_fields=['raw'],topk=topk)return "".join([item.fields['raw'] for item in rsp.output])
# query the top 1 results
question = '清华博士发生了什么?'
context = search_relevant_context(question, topk=1)
print(context)
2006-08-26 10:41:45
8月23日上午9时40分,京沪高速公路沧州服务区附近,一辆由北向南行驶的金杯面包车撞到高速公路护栏上,车上5名清华大学博士后研究人员及1名司机受伤,被紧急送往沧州二医院抢救。截至发稿时,仍有一名张姓博士后研究人员尚未脱离危险。
# initialize qwen 7B model
from modelscope import AutoModelForCausalLM, AutoTokenizer
from modelscope import GenerationConfigtokenizer = AutoTokenizer.from_pretrained("qwen/Qwen-7B-Chat", revision = 'v1.0.5',trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("qwen/Qwen-7B-Chat", revision = 'v1.0.5',device_map="auto", trust_remote_code=True, fp16=True).eval()
model.generation_config = GenerationConfig.from_pretrained("Qwen/Qwen-7B-Chat",revision = 'v1.0.5', trust_remote_code=True) # 可指定不同的生成长度、top_p等相关超参
问答
# define a prompt template for the vectorDB-enhanced LLM generation
def answer_question(question, context):prompt = f'''请基于```内的内容回答问题。"\```{context}\```我的问题是:{question}。'''history = Noneprint(prompt)response, history = model.chat(tokenizer, prompt, history=None) return response
# test the case on plain LLM without vectorDB enhancement
question = '清华博士发生了什么?'
answer = answer_question(question, '')
print(f'question: {question}\n' f'answer: {answer}')
请基于```内的内容回答问题。"
\```\```
我的问题是:清华博士发生了什么?。question: 清华博士发生了什么?
answer: 清华博士是指清华大学的博士研究生。作为一名AI语言模型,我无法获取个人的身份信息或具体事件,因此无法回答清华博士发生了什么。如果您需要了解更多相关信息,建议您查询相关媒体或官方网站。
# test the case with knowledge
context = search_relevant_context(question, topk=1)
answer = answer_question(question, context)
print(f'question: {question}\n' f'answer: {answer}')
请基于```内的内容回答问题。"
\```2006-08-26 10:41:45
8月23日上午9时40分,京沪高速公路沧州服务区附近,一辆由北向南行驶的金杯面包车撞到高速公路护栏上,车上5名清华大学博士后研究人员及1名司机受伤,被紧急送往沧州二医院抢救。截至发稿时,仍有一名张姓博士后研究人员尚未脱离危险。\```我的问题是:清华博士发生了什么?。question: 清华博士发生了什么?
answer: 8月23日上午9时40分,一辆由北向南行驶的金杯面包车撞到高速公路护栏上,车上5名清华大学博士后研究人员及1名司机受伤,被紧急送往沧州二医院抢救。
2024-03-24(日)
这篇关于RAG - QA + Qwen + dashscope的文章就介绍到这儿,希望我们推荐的文章对编程师们有所帮助!