README.md•2.02 kB
# RAG Tools
**RAG** tools:
- rag_Execute_Workflow - executes complete RAG pipeline (config setup, query storage, embedding generation, and semantic search)
**Configuration:**
The RAG system is fully configurable through `rag_config.yml`. You can customize:
- **Database locations** (query_db, model_db, vector_db)
- **Table names** (query_table, vector_table, model_table, etc.)
- **Model settings** (model_id, embedding dimensions)
- **Vector store metadata fields**
- **Embedding parameters** (vector length, column prefix, distance measure)
- **Retrieval settings** (default chunk count, maximum limits)
**Version Selection:**
The RAG tool supports two implementations:
- **BYOM (default)**: Uses ONNXEmbeddings for embedding generation
- **IVSM**: Uses IVSM functions for embedding generation
To switch between versions, edit `rag_config.yml`:
```yaml
version: 'byom' # Options: 'byom' or 'ivsm'
```
**Vector Store Compatibility:**
The system automatically adapts to your vector store schema. Configure your setup in `rag_config.yml`:
```yaml
# Database Configuration
databases:
query_db: "your_db"
vector_db: "your_vector_db"
# Table Configuration
tables:
vector_table: "your_vector_store_table"
# Model Configuration (adjust for different embedding models)
model:
model_id: "your-model-id"
# RAG Retrieval Configuration
retrieval:
default_k: 10 # Default number of chunks to retrieve
max_k: 50 # Maximum allowed chunks
# Embedding Configuration (change for different model dimensions)
embedding:
vector_length: 384 # Change based on your model
feature_columns: "[emb_0:emb_383]" # Adjust range accordingly
# Vector Store Schema
vector_store_schema:
metadata_fields_in_vector_store:
- "chunk_num"
- "doc_name"
# Add any other metadata columns from your vector store
```
The RAG tool supports two implementations that can be selected via configuration:
- rag_guidelines - guidelines for llm for rag workflow.
[Return to Main README](../../../../README.md)