2025最新版微软GraphRAG 2.0.0本地部署教程：基于Ollama快速构建知识图谱

一、前言

微软近期发布了知识图谱工具 GraphRAG 2.0.0，支持基于本地大模型（Ollama）快速构建知识图谱，显著提升了RAG（检索增强生成）的效果。本文手把手教你如何从零部署，并附踩坑记录和性能实测！

二、环境准备

1. 创建虚拟环境

推荐使用 Python 3.12.4（亲测兼容性较佳）：

  1. conda create -n graphrag200 python=3.12.4
  2. conda activate graphrag200

2. 拉取源码

建议通过Git下载最新代码（Windows用户需提前安装Git）：

  1. git clone https://github.com/microsoft/graphrag.git
  2. cd graphrag

    （附：若直接下载压缩包解压，解压完后需创建一个仓库，不然后续会报错）

        创建仓库方法：

  1. git init
  2. git add .
  3. git commit -m "Initial commit"

3. 安装依赖

一键安装所需依赖包：

pip install -e .

4. 创建输入文件夹

用于存放待处理的文档（Windows可以直接手动创建）：

mkdir -p ./graphrag_ollama/input

将数据集放入input目录即可。

三、关键配置修改

1. 初始化项目

执行初始化命令（注意与旧版参数不同）：

python -m graphrag init --root ./graphrag_ollama

2. 修改settings.yaml

核心配置项（需按需调整）：

  • 模型设置：使用Ollama本地模型

 注意修改一下圈出的几个地方

测试小文件时，建议把chunks改小：

 修改结果如下：

    1. ### This config file contains required core defaults that must be set, along with a handful of common optional settings.
    2. ### For a full list of available settings, see https://microsoft.github.io/graphrag/config/yaml/
    3.  
    4. ### LLM settings ###
    5. ## There are a number of settings to tune the threading and token limits for LLM calls - check the docs.
    6.  
    7. models:
    8.   default_chat_model:
    9.     type: openai_chat # or azure_openai_chat
    10.     api_base: http://192.168.0.167:11434/v1
    11.     # api_version: 2024-05-01-preview
    12.     auth_type: api_key # or azure_managed_identity
    13.     api_key: ${GRAPHRAG_API_KEY} # set this in the generated .env file
    14.     # audience: "https://cognitiveservices.azure.com/.default"
    15.     # organization: <organization_id>
    16.     model: deepseek-r1:32b
    17.     # deployment_name: <azure_model_deployment_name>
    18.     encoding_model: cl100k_base # automatically set by tiktoken if left undefined
    19.     model_supports_json: true # recommended if this is available for your model.
    20.     concurrent_requests: 25 # max number of simultaneous LLM requests allowed
    21.     async_mode: threaded # or asyncio
    22.     retry_strategy: native
    23.     max_retries: -1                   # set to -1 for dynamic retry logic (most optimal setting based on server response)
    24.     tokens_per_minute: 0              # set to 0 to disable rate limiting
    25.     requests_per_minute: 0            # set to 0 to disable rate limiting
    26.   default_embedding_model:
    27.     type: openai_embedding # or azure_openai_embedding
    28.     api_base: http://192.168.0.167:11434/v1
    29.     # api_version: 2024-05-01-preview
    30.     auth_type: api_key # or azure_managed_identity
    31.     api_key: ${GRAPHRAG_API_KEY}
    32.     # audience: "https://cognitiveservices.azure.com/.default"
    33.     # organization: <organization_id>
    34.     model: bge-m3:latest
    35.     # deployment_name: <azure_model_deployment_name>
    36.     encoding_model: cl100k_base # automatically set by tiktoken if left undefined
    37.     model_supports_json: true # recommended if this is available for your model.
    38.     concurrent_requests: 25 # max number of simultaneous LLM requests allowed
    39.     async_mode: threaded # or asyncio
    40.     retry_strategy: native
    41.     max_retries: -1                   # set to -1 for dynamic retry logic (most optimal setting based on server response)
    42.     tokens_per_minute: 0              # set to 0 to disable rate limiting
    43.     requests_per_minute: 0            # set to 0 to disable rate limiting
    44.  
    45. vector_store:
    46.   default_vector_store:
    47.     type: lancedb
    48.     db_uri: output\lancedb
    49.     container_name: default
    50.     overwrite: True
    51.  
    52. embed_text:
    53.   model_id: default_embedding_model
    54.   vector_store_id: default_vector_store
    55.  
    56. ### Input settings ###
    57.  
    58. input:
    59.   type: file # or blob
    60.   file_type: text # or csv
    61.   base_dir: "input"
    62.   file_encoding: utf-8
    63.   file_pattern: ".*\\.txt$$"
    64.  
    65. chunks:
    66.   size: 200
    67.   overlap: 50
    68.   group_by_columns: [id]
    69.  
    70. ### Output settings ###
    71. ## If blob storage is specified in the following four sections,
    72. ## connection_string and container_name must be provided
    73.  
    74. cache:
    75.   type: file # [file, blob, cosmosdb]
    76.   base_dir: "cache"
    77.  
    78. reporting:
    79.   type: file # [file, blob, cosmosdb]
    80.   base_dir: "logs"
    81.  
    82. output:
    83.   type: file # [file, blob, cosmosdb]
    84.   base_dir: "output"
    85.  
    86. ### Workflow settings ###
    87.  
    88. extract_graph:
    89.   model_id: default_chat_model
    90.   prompt: "prompts/extract_graph.txt"
    91.   entity_types: [organization,person,geo,event]
    92.   max_gleanings: 1
    93.  
    94. summarize_descriptions:
    95.   model_id: default_chat_model
    96.   prompt: "prompts/summarize_descriptions.txt"
    97.   max_length: 500
    98.  
    99. extract_graph_nlp:
    100.   text_analyzer:
    101.     extractor_type: regex_english # [regex_english, syntactic_parser, cfg]
    102.  
    103. extract_claims:
    104.   enabled: false
    105.   model_id: default_chat_model
    106.   prompt: "prompts/extract_claims.txt"
    107.   description: "Any claims or facts that could be relevant to information discovery."
    108.   max_gleanings: 1
    109.  
    110. community_reports:
    111.   model_id: default_chat_model
    112.   graph_prompt: "prompts/community_report_graph.txt"
    113.   text_prompt: "prompts/community_report_text.txt"
    114.   max_length: 2000
    115.   max_input_length: 8000
    116.  
    117. cluster_graph:
    118.   max_cluster_size: 10
    119.  
    120. embed_graph:
    121.   enabled: false # if true, will generate node2vec embeddings for nodes
    122.  
    123. umap:
    124.   enabled: false # if true, will generate UMAP embeddings for nodes (embed_graph must also be enabled)
    125.  
    126. snapshots:
    127.   graphml: false
    128.   embeddings: false
    129.  
    130. ### Query settings ###
    131. ## The prompt locations are required here, but each search method has a number of optional knobs that can be tuned.
    132. ## See the config docs: https://microsoft.github.io/graphrag/config/yaml/#query
    133.  
    134. local_search:
    135.   chat_model_id: default_chat_model
    136.   embedding_model_id: default_embedding_model
    137.   prompt: "prompts/local_search_system_prompt.txt"
    138.  
    139. global_search:
    140.   chat_model_id: default_chat_model
    141.   map_prompt: "prompts/global_search_map_system_prompt.txt"
    142.   reduce_prompt: "prompts/global_search_reduce_system_prompt.txt"
    143.   knowledge_prompt: "prompts/global_search_knowledge_system_prompt.txt"
    144.  
    145. drift_search:
    146.   chat_model_id: default_chat_model
    147.   embedding_model_id: default_embedding_model
    148.   prompt: "prompts/drift_search_system_prompt.txt"
    149.   reduce_prompt: "prompts/drift_search_reduce_prompt.txt"
    150.  
    151. basic_search:
    152.   chat_model_id: default_chat_model
    153.   embedding_model_id: default_embedding_model
    154.   prompt: "prompts/basic_search_system_prompt.txt"

    四、构建知识图谱

    执行索引命令（算力警告：亲测4090-24G显卡处理2万字需3小时）：

    python -m graphrag index --root ./graphrag_ollama

    五、知识图谱查询

    支持多种查询方式，按需选择：

  • 方法命令示例用途
    全局查询python -m graphrag query --method global --query "知识图谱定义"跨文档综合分析
    局部查询python -m graphrag query --method local --query "知识图谱定义"单文档精准检索
    DRIFT查询python -m graphrag query --method drift --query "知识图谱定义"动态漂移分析
    基础查询python -m graphrag query --method basic --query "知识图谱定义"传统RAG检索

六、注意事项

  1. 模型路径：确保Ollama服务已启动，且模型名称与配置一致（如deepseek-r1:32b需提前拉取）。

  2. 算力需求：小规模数据集建议使用GPU加速，CPU模式耗时可能成倍增加。

  3. 文件编码：输入文档需为UTF-8编码，否则可能报错。

  4. 配置备份：修改settings.yaml前建议备份原始文件。

七、总结

GraphRAG 2.0.0大幅优化了知识图谱的构建效率，结合本地模型可实现隐私安全的行业级应用。若遇到部署问题，欢迎在评论区留言交流！

相关资源：

 GraphRAG GitHub仓库

Ollama模型库

原创声明：本文为作者原创，未经授权禁止转载。如需引用请联系作者。

点赞关注，技术不迷路！ 👍
你的支持是我更新的最大动力！ ⚡

注：本文转载自blog.csdn.net的Sindy_he的文章"https://blog.csdn.net/m0_54356251/article/details/146074188"。版权归原作者所有，此博客不拥有其著作权，亦不承担相应法律责任。如有侵权，请联系我们删除。
复制链接
相关推荐
发表评论

评论记录：

未查询到任何数据！