Greennode 集成 - 以编程方式使用文档

>GreenNode 是全球 AI 解决方案提供商，也是 **NVIDIA 优选合作伙伴**，提供从基础设施到应用程序的全栈 AI 能力，服务于美国、中东和北非以及亚太地区的企业。在 **世界级基础设施** （LEED 金级、TIA‑942、Uptime 三级）上运行，GreenNode 为企业、初创公司和研究人员提供全面的 AI 服务套件

本指南提供有关开始使用 GreenNodeRerank 检索器的详细说明。它使您能够使用内置连接器或集成您自己的数据源执行文档搜索，利用 GreenNode 的重排序功能来提高相关性。

集成详情

提供商: GreenNode 无服务器 AI
模型类型：重排序模型
主要用例：根据语义相关性对搜索结果进行重排序
可用模型：包含 BAAI/bge-reranker-v2-m3 以及其他高性能重排序模型
评分：返回用于根据查询匹配度重新排序文档候选的相关性分数

设置

要访问 GreenNode 模型，您需要创建一个 GreenNode 账户，获取 API 密钥，并安装 langchain-greennode 集成包。

凭证

前往此页面注册 GreenNode AI 平台并生成 API 密钥。完成此操作后，设置 GREENNODE_API_KEY 环境变量：

if not os.getenv("GREENNODE_API_KEY"):
    os.environ["GREENNODE_API_KEY"] = getpass.getpass("Enter your GreenNode API key: ")

如果您想从单个查询获取自动追踪，还可以设置您的 LangSmith API 密钥，方法是在下面取消注释：

os.environ["LANGSMITH_API_KEY"] = getpass.getpass("Enter your LangSmith API key: ")
os.environ["LANGSMITH_TRACING"] = "true"

安装

此检索器位于 langchain-greennode package:

pip install -qU langchain-greennode

实例化

可以使用可选参数（API 密钥和模型名称）实例化 GreenNodeRerank 类：

from langchain_greennode import GreenNodeRerank

# Initialize the embeddings model
reranker = GreenNodeRerank(
    # api_key="YOUR_API_KEY",  # You can pass the API key directly
    model="BAAI/bge-reranker-v2-m3",  # The default embedding model
    top_n=3,
)

用法

对搜索结果进行重排序

重排序模型通过基于语义相关性细化和重新排序初始搜索结果来增强检索增强生成（RAG）工作流程。以下示例演示了如何将 GreenNodeRerank 与基础检索器集成以提高检索文档的质量。

from langchain_classic.retrievers.contextual_compression import ContextualCompressionRetriever
from langchain_community.vectorstores import FAISS
from langchain_core.documents import Document
from langchain_greennode import GreenNodeEmbeddings

# Initialize the embeddings model
embeddings = GreenNodeEmbeddings(
    # api_key="YOUR_API_KEY",  # You can pass the API key directly
    model="BAAI/bge-m3"  # The default embedding model
)

# Prepare documents (finance/economics domain)
docs = [
    Document(
        page_content="Inflation represents the rate at which the general level of prices for goods and services rises"
    ),
    Document(
        page_content="Central banks use interest rates to control inflation and stabilize the economy"
    ),
    Document(
        page_content="Cryptocurrencies like Bitcoin operate on decentralized blockchain networks"
    ),
    Document(
        page_content="Stock markets are influenced by corporate earnings, investor sentiment, and economic indicators"
    ),
]

# Create a vector store and a base retriever
vector_store = FAISS.from_documents(docs, embeddings)
base_retriever = vector_store.as_retriever(search_kwargs={"k": 4})


rerank_retriever = ContextualCompressionRetriever(
    base_compressor=reranker, base_retriever=base_retriever
)

# Perform retrieval with reranking
query = "How do central banks fight rising prices?"
results = rerank_retriever.get_relevant_documents(query)

results

/var/folders/bs/g52lln652z11zjp98qf9wcy40000gn/T/ipykernel_96362/2544494776.py:41: LangChainDeprecationWarning: The method `BaseRetriever.get_relevant_documents` was deprecated in langchain-core 0.1.46 and will be removed in 1.0. Use :meth:`~invoke` instead.
  results = rerank_retriever.get_relevant_documents(query)

[Document(metadata={'relevance_score': 0.125}, page_content='Central banks use interest rates to control inflation and stabilize the economy'),
 Document(metadata={'relevance_score': 0.004913330078125}, page_content='Inflation represents the rate at which the general level of prices for goods and services rises'),
 Document(metadata={'relevance_score': 1.6689300537109375e-05}, page_content='Cryptocurrencies like Bitcoin operate on decentralized blockchain networks')]

直接使用

类可以独立使用，根据相关性分数对检索到的文档进行重排序。当主要检索步骤（例如关键词或向量搜索）返回大量候选文档，而需要使用更复杂的语义理解来细化结果的次级模型时，此功能特别有用。该类接受一个查询和一组候选文档，并返回基于预测相关性重新排序的列表。 GreenNodeRerank

test_documents = [
    Document(
        page_content="Carson City is the capital city of the American state of Nevada."
    ),
    Document(
        page_content="Washington, D.C. (also known as simply Washington or D.C.) is the capital of the United States."
    ),
    Document(
        page_content="Capital punishment has existed in the United States since beforethe United States was a country."
    ),
    Document(
        page_content="The Commonwealth of the Northern Mariana Islands is a group of islands in the Pacific Ocean. Its capital is Saipan."
    ),
]

test_query = "What is the capital of the United States?"
results = reranker.rerank(test_documents, test_query)
results

[{'index': 1, 'relevance_score': 1.0},
 {'index': 0, 'relevance_score': 0.01165771484375},
 {'index': 3, 'relevance_score': 0.0012054443359375}]

在链中使用

GreenNodeRerank 可在 LangChain RAG 管道中无缝工作。以下是使用 GreenNodeRerank 创建简单 RAG 链的示例：

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_greennode import ChatGreenNode

# Initialize LLM
llm = ChatGreenNode(model="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B")

# Create a prompt template
prompt = ChatPromptTemplate.from_template(
    """
Answer the question based only on the following context:

Context:
{context}

Question: {question}
"""
)


# Format documents function
def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)


# Create RAG chain
rag_chain = (
    {"context": rerank_retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# Run the chain
answer = rag_chain.invoke("How do central banks fight rising prices?")
answer

'\n\nCentral banks combat rising prices, or inflation, by adjusting interest rates. By raising interest rates, they increase the cost of borrowing, which discourages spending and investment. This reduction in demand helps slow down the rate of price increases, thereby controlling inflation and contributing to economic stability.'

API 参考文档

有关 GreenNode Serverless AI API 的更多详细信息，请访问 GreenNode Serverless AI 文档.