当前位置：首页 > news >正文

【简单学习】llamaindex环境搭建以及构建RAG

news 2025/8/20 2:04:11

安装

1、创建一个创建虚拟环境

conda create -n llamaindex python==3.12 -y
conda activate llamaindex

2、安装 llama-index
注意llama-index每一个版本都互不兼容

 ###目前版本是0.12.32
pip install llama-index -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install llama-index-embeddings-huggingface -i https://pypi.tuna.tsinghua.edu.cn/simplepip install llama-index-llms-huggingface -i https://pypi.tuna.tsinghua.edu.cn/simple

3、代码示例

from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core import Settings,SimpleDirectoryReader,VectorStoreIndex
from llama_index.llms.huggingface import HuggingFaceLLM#初始化一个HuggingFaceEmbedding对象，用于将文本转换为向量表示
embed_model = HuggingFaceEmbedding(#指定了一个预训练的sentence-transformer模型的路径model_name=r"D:model\sentence-transformers\paraphrase-multilingual-MiniLM-L12-v2"
)#将创建的嵌入模型赋值给全局设置的embed_model属性，这样在后续的索引构建过程中，就会使用这个模型
Settings.embed_model = embed_model#使用HuggingFaceLLM加载本地大模型
llm = HuggingFaceLLM(model_name=r"D:models\Qwen\Qwen2___5-1___5B-Instruct",tokenizer_name=r"D:models\Qwen\Qwen2___5-1___5B-Instruct",model_kwargs={"trust_remote_code":True},tokenizer_kwargs={"trust_remote_code":True,"use_sliding_window_attention": False})
#设置全局的llm属性，这样在索引查询时会使用这个模型。
Settings.llm = llm#从指定目录读取文档，将数据加载到内存
documents = SimpleDirectoryReader(r"D:\data").load_data()
# print(documents)
#创建一个VectorStoreIndex,并使用之前加载的文档来构建向量索引
#此索引将文档转换为向量，并存储这些向量（内存）以便于快速检索
index = VectorStoreIndex.from_documents(documents)
#创建一个查询引擎，这个引擎可以接收查询并返回相关文档的响应。
query_engine = index.as_query_engine()
rsp = query_engine.query("xtuner是什么？")
print(rsp)

4、运行代码报错如下

D:\envs\llamaindex\python.exe "D:\Program Files\python\PycharmProjects\AiStudyProject\demo17\test02.py" 
[nltk_data] Error loading stopwords: <urlopen error [Errno 11004]
[nltk_data]     getaddrinfo failed>
[nltk_data] Error loading punkt: <urlopen error [Errno 11004]
[nltk_data]     getaddrinfo failed>
Sliding Window Attention is enabled but not implemented for `sdpa`; unexpected results may be encountered.
[nltk_data] Error loading stopwords: <urlopen error [Errno 11004]
[nltk_data]     getaddrinfo failed>
[nltk_data] Error loading punkt_tab: <urlopen error [Errno 11004]
[nltk_data]     getaddrinfo failed>
Traceback (most recent call last):File "D:\envs\llamaindex\Lib\site-packages\nltk\corpus\util.py", line 84, in __loadroot = nltk.data.find(f"{self.subdir}/{zip_name}")^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\nltk\data.py", line 579, in findraise LookupError(resource_not_found)
LookupError: 
**********************************************************************Resource stopwords not found.Please use the NLTK Downloader to obtain the resource:>>> import nltk>>> nltk.download('stopwords')For more information see: https://www.nltk.org/data.htmlAttempted to load corpora/stopwords.zip/stopwords/Searched in:- 'C:\\Users\\fengxinzi/nltk_data'- 'D:\\envs\\llamaindex\\nltk_data'- 'D:\\envs\\llamaindex\\share\\nltk_data'- 'D:\\envs\\llamaindex\\lib\\nltk_data'- 'C:\\Users\\fengxinzi\\AppData\\Roaming\\nltk_data'- 'C:\\nltk_data'- 'D:\\nltk_data'- 'E:\\nltk_data'- 'D:\\envs\\llamaindex\\Lib\\site-packages\\llama_index\\core\\_static/nltk_cache'
**********************************************************************During handling of the above exception, another exception occurred:Traceback (most recent call last):File "D:\Program Files\python\PycharmProjects\AiStudyProject\demo17\test02.py", line 30, in <module>index = VectorStoreIndex.from_documents(documents)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\indices\base.py", line 112, in from_documentsnodes = run_transformations(^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\ingestion\pipeline.py", line 100, in run_transformationsnodes = transform(nodes, **kwargs)^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\instrumentation\dispatcher.py", line 322, in wrapperresult = func(*args, **kwargs)^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\interface.py", line 193, in __call__return self.get_nodes_from_documents(nodes, **kwargs)  # type: ignore^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\interface.py", line 165, in get_nodes_from_documentsnodes = self._parse_nodes(documents, show_progress=show_progress, **kwargs)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\instrumentation\dispatcher.py", line 322, in wrapperresult = func(*args, **kwargs)^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\interface.py", line 260, in _parse_nodessplits = self.split_text_metadata_aware(^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\instrumentation\dispatcher.py", line 322, in wrapperresult = func(*args, **kwargs)^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\text\sentence.py", line 173, in split_text_metadata_awarereturn self._split_text(text, chunk_size=effective_chunk_size)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\text\sentence.py", line 190, in _split_textsplits = self._split(text, chunk_size)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\text\sentence.py", line 211, in _splittext_splits_by_fns, is_sentence = self._get_splits_by_fns(text)^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\text\sentence.py", line 313, in _get_splits_by_fnssplits = split_fn(text)^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\node_parser\text\utils.py", line 84, in <lambda>text, globals_helper.punkt_tokenizer^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\llama_index\core\utils.py", line 117, in punkt_tokenizerself._stopwords = stopwords.words("english")^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\nltk\corpus\util.py", line 120, in __getattr__self.__load()File "D:\envs\llamaindex\Lib\site-packages\nltk\corpus\util.py", line 86, in __loadraise eFile "D:\envs\llamaindex\Lib\site-packages\nltk\corpus\util.py", line 81, in __loadroot = nltk.data.find(f"{self.subdir}/{self.__name}")^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^File "D:\envs\llamaindex\Lib\site-packages\nltk\data.py", line 579, in findraise LookupError(resource_not_found)
LookupError: 
**********************************************************************Resource stopwords not found.Please use the NLTK Downloader to obtain the resource:>>> import nltk>>> nltk.download('stopwords')For more information see: https://www.nltk.org/data.htmlAttempted to load corpora/stopwordsSearched in:- 'C:\\Users\\fengxinzi/nltk_data'- 'D:\\envs\\llamaindex\\nltk_data'- 'D:\\envs\\llamaindex\\share\\nltk_data'- 'D:\\envs\\llamaindex\\lib\\nltk_data'- 'C:\\Users\\fengxinzi\\AppData\\Roaming\\nltk_data'- 'C:\\nltk_data'- 'D:\\nltk_data'- 'E:\\nltk_data'- 'D:\\envs\\llamaindex\\Lib\\site-packages\\llama_index\\core\\_static/nltk_cache'
**********************************************************************Process finished with exit code 1

错误表明 NLTK 仍无法加载 stopwords 和 punkt 资源。
我的nltk版本如下：
在这里插入图片描述

5、解决方案：手动下载并放置资源文件

下载资源包：
- 访问 NLTK 下载页面，下载以下文件：
  - stopwords.zip
  - punkt.zip

stopwords下载点击直达
punkt下载点击直达

解压到 NLTK 的搜索路径：
根据错误信息，优先将文件解压到以下路径：

D:\envs\llamaindex\nltk_data\corpora\    # 解压 stopwords.zip 到此路径
D:\envs\llamaindex\nltk_data\tokenizers\ # 解压 punkt.zip 到此路径

如果路径不存在，手动创建文件夹。

验证资源加载：
在 Python 中运行以下代码，检查资源是否能正确加载：
```
import nltk# 尝试加载资源
print(nltk.data.find('corpora/stopwords'))
print(nltk.data.find('tokenizers/punkt'))
```
输出路径（如 D:\envs\llamaindex\nltk_data\corpora\stopwords），说明配置成功。

6、再次测试成功

from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core import Settings,SimpleDirectoryReader,VectorStoreIndex
from llama_index.llms.huggingface import HuggingFaceLLM
import nltk
# 下载 NLTK 资源
# 尝试加载资源
print(nltk.data.find('corpora/stopwords'))
print(nltk.data.find('tokenizers/punkt'))#初始化一个HuggingFaceEmbedding对象，用于将文本转换为向量表示
embed_model = HuggingFaceEmbedding(#指定了一个预训练的sentence-transformer模型的路径model_name=r"D:\model\sentence-transformers\paraphrase-multilingual-MiniLM-L12-v2"
)#将创建的嵌入模型赋值给全局设置的embed_model属性，这样在后续的索引构建过程中，就会使用这个模型
Settings.embed_model = embed_model#使用HuggingFaceLLM加载本地大模型
llm = HuggingFaceLLM(model_name=r"D:\models\Qwen\Qwen2___5-1___5B-Instruct",tokenizer_name=r"D:models\Qwen\Qwen2___5-1___5B-Instruct",model_kwargs={"trust_remote_code":True},tokenizer_kwargs={"trust_remote_code":True,"use_sliding_window_attention": False})
#设置全局的llm属性，这样在索引查询时会使用这个模型。
Settings.llm = llm#从指定目录读取文档，将数据加载到内存
documents = SimpleDirectoryReader(r"D:\data").load_data()
# print(documents)
#创建一个VectorStoreIndex,并使用之前加载的文档来构建向量索引
#此索引将文档转换为向量，并存储这些向量（内存）以便于快速检索
index = VectorStoreIndex.from_documents(documents)
#创建一个查询引擎，这个引擎可以接收查询并返回相关文档的响应。
query_engine = index.as_query_engine()
rsp = query_engine.query("xtuner是什么？")
print(rsp)