如果你在 Colab 中打开这个 Notebook，可能需要先安装 LlamaIndex 🦙。

%pip install llama-index-vector-stores-pinecone "transformers[torch]"

创建 Pinecone 索引

from pinecone import Pinecone, ServerlessSpec

import os

os.environ["PINECONE_API_KEY"] = "..."
os.environ["OPENAI_API_KEY"] = "sk-..."

api_key = os.environ["PINECONE_API_KEY"]

pc = Pinecone(api_key=api_key)

# delete if needed
pc.delete_index("quickstart")

# dimensions are for text-embedding-ada-002
# NOTE: needs dotproduct for hybrid search

pc.create_index(
    name="quickstart",
    dimension=1536,
    metric="dotproduct",
    spec=ServerlessSpec(cloud="aws", region="us-east-1"),
)

# If you need to create a PodBased Pinecone index, you could alternatively do this:
#
# from pinecone import Pinecone, PodSpec
#
# pc = Pinecone(api_key='xxx')
#
# pc.create_index(
#    name='my-index',
#    dimension=1536,
#    metric='cosine',
#    spec=PodSpec(
#      environment='us-east1-gcp',
#      pod_type='p1.x1',
#      pods=1
#    )
# )
#

pinecone_index = pc.Index("quickstart")

下载数据

!mkdir -p 'data/paul_graham/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/examples/data/paul_graham/paul_graham_essay.txt' -O 'data/paul_graham/paul_graham_essay.txt'

加载文档，构建 PineconeVectorStore

当 add_sparse_vector=True 时，PineconeVectorStore 将为每个文档计算稀疏向量。

默认情况下，它使用简单的 token 频率来生成稀疏向量。但你也可以指定自定义稀疏嵌入模型。

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.vector_stores.pinecone import PineconeVectorStore
from IPython.display import Markdown, display

# load documents
documents = SimpleDirectoryReader("./data/paul_graham/").load_data()

# set add_sparse_vector=True to compute sparse vectors during upsert
from llama_index.core import StorageContext

if "OPENAI_API_KEY" not in os.environ:
    raise EnvironmentError(f"Environment variable OPENAI_API_KEY is not set")

vector_store = PineconeVectorStore(
    pinecone_index=pinecone_index,
    add_sparse_vector=True,
)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
To disable this warning, you can either:
  - Avoid using `tokenizers` before the fork if possible
  - Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)

Upserted vectors:   0%|          | 0/22 [00:00<?, ?it/s]

查询索引

可能需要等待一两分钟让索引就绪

# set Logging to DEBUG for more detailed outputs
query_engine = index.as_query_engine(vector_store_query_mode="hybrid")
response = query_engine.query("What happened at Viaweb?")

display(Markdown(f"<b>{response}</b>"))

Paul Graham 创办 Viaweb 是因为他需要钱。随着公司发展，他意识到自己不想经营一家大公司，并决定将愿景的一个子集构建为开源项目。最终，Viaweb 在 1998 年夏天被 Yahoo 收购，这对 Paul Graham 来说是巨大的解脱。

更改稀疏嵌入模型

%pip install llama-index-sparse-embeddings-fastembed

# Clear the vector store
vector_store.clear()

from llama_index.sparse_embeddings.fastembed import FastEmbedSparseEmbedding

sparse_embedding_model = FastEmbedSparseEmbedding(
    model_name="prithivida/Splade_PP_en_v1"
)

vector_store = PineconeVectorStore(
    pinecone_index=pinecone_index,
    add_sparse_vector=True,
    sparse_embedding_model=sparse_embedding_model,
)

Fetching 5 files:   0%|          | 0/5 [00:00<?, ?it/s]

index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

Upserted vectors:   0%|          | 0/22 [00:00<?, ?it/s]

等待一分钟让内容上传完成..

response = query_engine.query("What happened at Viaweb?")
display(Markdown(f"<b>{response}</b>"))

Paul Graham 创办 Viaweb 是因为他需要钱。他招募了一个团队来构建软件和服务，重点是创建应用构建器和网络基础设施。然而，夏天过半时，Paul 意识到自己不想经营一家大公司，并决定将重点转向将项目的一个子集构建为开源项目。这促成了 Lisp 新方言 Arc 的开发。最终，Viaweb 在 1998 年夏天被出售给 Yahoo，这让 Paul Graham 松了一口气，并使他得以过渡到人生的新阶段。