Skip to main content
This quickstart guide walks you through setting up and running a basic RAG optimization pipeline. You’ll learn how to load dataset files, generate question–answer pairs, define a RAG configuration with tunable hyperparameters and models, and execute the optimization workflow. You can also try it out directly in this Notebook
1

Install Requirements

Install RAGOPT via pip:
pip install rag-opt
2

Prepare the RAG Configuration File

Create a configuration file named rag_config.yaml in your project directory. This file defines the search space for your optimization. You can also copy a sample configuration from the repository RAG Parameters Config YAML File
All available Hugging Face models and inference providers can be found at https://hf.co/settings/inference-providers.
Example:
# Minimal RAG Configuration Example

chunk_size:
  bounds: [512, 1024]
  dtype: int

max_tokens:
  bounds: [256, 512]
  dtype: int

chunk_overlap:
  bounds: [0, 200]
  dtype: int

temperature:
  bounds: [0.0, 1.0]
  dtype: float

vector_store:
  choices:
    faiss: {}
    pinecone:
      api_key: "YOUR_PINECONE_API_KEY"
      index_name: "your-index-name"

search_type:
  choices: ["similarity", "mmr"]

embedding:
  choices:
    huggingface:
      models:
        - "all-MiniLM-L6-v2"
    openai:
      api_key: "YOUR_OPENAI_API_KEY"
      models:
        - "text-embedding-ada-002"

k:
  bounds: [1, 10]
  dtype: int

use_reranker: false

llm:
  choices:
    openai:
      api_key: "YOUR_OPENAI_API_KEY"
      models:
        - "gpt-3.5-turbo"
    huggingface:
      models:
        - "gpt2-medium"
3

Generate the Dataset

Generate a set of question–answer pairs to be used for optimizing the RAG pipeline.You can use any LLM you prefer from Available Chat Models
from langchain.chat_models import init_chat_model
from rag_opt.rag import DatasetGenerator

# Initialize the language model
llm = init_chat_model(
    model="gpt-3.5-turbo",
    model_provider="openai",
    api_key="sk-***"
)

# Generate training data
data_gen = DatasetGenerator(llm,dataset_path="./data")
dataset = data_gen.generate(3)  # Generates 3 training samples
dataset.to_json("./rag_dataset.json")
4

Run Optimization

Run the optimization process using your generated dataset and configuration file.
from rag_opt.dataset import TrainDataset
from rag_opt.optimizer import Optimizer

# Load the training dataset
train_dataset = TrainDataset.from_json("rag_dataset.json")

# Initialize the optimizer
optimizer = Optimizer(
train_dataset=train_dataset,
config_path="rag_config.yaml",
acquisition_functions=['qEHVI','Random'], # ['qEHVI', 'qNEHVI', 'qNParEGO', 'Random']
verbose=True
)

# Run optimization
best_config = optimizer.optimize(n_trials=3, best_one=True)  # Increase n_trials for better results
best_config.to_json()

​
Next Steps

  • Explore the Optimization Workflow for a deeper dive into the pipeline.
  • Review Metrics Overview for advanced customization options.
  • Build and experiment with your own RAG pipeline for production use.