Configuration Guide
4 minute read
Config
python -m NodeRAG.build -f path/to/main_foulder
When you first use this command, it will create Node_config.yaml file in the main_folder directory.
You need to modify the config file according to the following instructions (add API and service provider) to ensure that NodeRAG can access the correct API.
#==============================================================================
# AI Model Configuration
#==============================================================================
model_config:
service_provider: openai # AI service provider (e.g., openai, azure)
model_name: gpt-4o-mini # Model name for text generation
api_keys: # Your API key (optional)
temperature: 0 # Temperature parameter for text generation
max_tokens: 10000 # Maximum tokens to generate
rate_limit: 40 # API rate limit (requests per second)
embedding_config:
service_provider: openai_embedding # Embedding service provider
embedding_model_name: text-embedding-3-small # Model name for text embeddings
api_keys: # Your API key (optional)
rate_limit: 20 # Rate limit for embedding requests
#==============================================================================
# Document Processing Configuration
#==============================================================================
config:
# Basic Settings
main_folder: C:\Users\Terry_Xu\Desktop\HP # Root folder for document processing
language: English # Document processing language
docu_type: mixed # Document type (mixed, pdf, txt, etc.)
# Chunking Settings
chunk_size: 1048 # Size of text chunks for processing
embedding_batch_size: 50 # Batch size for embedding processing
# UI Settings
use_tqdm: false # Enable/disable progress bars
use_rich: true # Enable/disable rich text formatting
# HNSW Index Settings
space: l2 # Distance metric for HNSW (l2, cosine)
dim: 1536 # Embedding dimension (must match embedding model)
m: 50 # Number of connections per layer in HNSW
ef: 200 # Size of dynamic candidate list in HNSW
m0: # Number of bi-directional links in HNSW
# Summary Settings
Hcluster_size: 39 # Number of clusters for high-level element matching
# Search Server Settings
url: '127.0.0.1' # Server URL for search service
port: 5000 # Server port number
unbalance_adjust: true # Enable adjustment for unbalanced data
cross_node: 10 # Number of cross nodes to return
Enode: 10 # Number of entity nodes to return
Rnode: 30 # Number of relationship nodes to return
Hnode: 10 # Number of high-level nodes to return
HNSW_results: 10 # Number of HNSW search results
similarity_weight: 1 # Weight for similarity in personalized PageRank
accuracy_weight: 1 # Weight for accuracy in personalized PageRank
ppr_alpha: 0.5 # Damping factor for personalized PageRank
ppr_max_iter: 2 # Maximum iterations for personalized PageRank
Configuration
service provider
We currently support model service providers such as openai and genmini. For embedding models, we support openai_embedding and gemini embedding. We strongly recommend using openai as the model provider. This is because openai’s structure decoding functionality can effectively follow structured decomposition prompts to generate heterogeneous graphs, providing great stability. Unfortunately, the current gemini models and other models (eg. DeepSeek), even the latest gemini 2.0, have shown instability in structured outputs. We discussed this issue in this blog.
Available Models
When you choose a service provider, you can use all the currently available models they offer.
For example, if you choose openai as the service provider, you can use all the models listed in the latest models on their website.
Here are some examples of available models for openai:
- gpt-4o-mini
- gpt-4
- gpt-4o
For embedding models, openai_embedding offers:
- text-embedding-3-small
- text-embedding-3-large
Please refer to the Embeddings for the most up-to-date list of available models.
Rate Limit
OpenAI accounts are subject to rate limits, including restrictions on the number of requests and tokens per minute. You can adjust the request rates for both the embedding model and the language model to control usage spikes by setting appropriate rate limits.
For more information on account tiers and specific rate limits, see the OpenAI Rate Limits Guide.
Language
Currently, English and Chinese are supported. To add support for additional languages, modify the prompt-related text and update the language return field in promanager accordingly. See Prompt Fine-Tuning for more details.
Embedding Dimension
dim represents the number of dimensions in the embedding vector. Please ensure this value matches the embedding model you are using. For example, OpenAI’s embedding dimension is 1536, while Gemini’s embedding dimension is 768.
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.