IntelLabs/RAG-FiT

Framework for enhancing LLMs for RAG tasks using fine-tuning.

What it solves

RAG-FiT (formerly RAG Foundry) addresses the challenge of improving how Large Language Models (LLMs) utilize external information. It provides a structured framework to fine-tune models specifically for Retrieval Augmented Generation (RAG) tasks, moving beyond simple prompt engineering to optimize the model's internal ability to process and reason over retrieved data.

How it works

The library is organized into four modular components that manage the entire RAG lifecycle:

  1. Dataset Creation: Processes raw data into RAG-augmented datasets, including retrieval, aggregation, and prompt formatting.
  2. Training: Uses Parameter-Efficient Fine-Tuning (PEFT) and TRL to train models on these augmented datasets.
  3. Inference: Generates predictions using the trained models.
  4. Evaluation: Assesses performance using a suite of RAG-specific metrics like RAGAS, BERTScore, and custom logic.

The framework uses Hydra for configuration-as-code, allowing users to manage complex experiments through hierarchical configuration files and CLI overrides.

Who it’s for

This tool is designed for researchers and developers who need to experiment with RAG configurations and improve LLM performance on knowledge-intensive tasks through fine-tuning.

Highlights

  • Modular architecture covering the full pipeline from data preparation to evaluation.
  • Native support for PEFT (Parameter-Efficient Fine-Tuning) to reduce training resource requirements.
  • Extensive evaluation suite supporting RAGAS, Deepeval, and custom metrics.
  • Configuration-driven workflow using Hydra, enabling reproducible experiments and integration with tools like SLURM and Ray.

Related

  • Project
    agentscope-ai/Trinity-RFT

    A general-purpose framework for reinforcement fine-tuning of LLMs that decouples exploration, training, and data buffering to support diverse RL modes and agentic workflows.

  • Project
    huggingface/peft

    A library providing state-of-the-art parameter-efficient fine-tuning methods to adapt large pretrained models using minimal compute and storage.

  • Project
    NovaSearch-Team/RAG-Retrieval

    A comprehensive toolkit for training, fine-tuning, and distilling RAG retrieval models, including embedding models, late interaction models, and rerankers.

  • Project
    data-prep-kit/data-prep-kit

    A scalable toolkit for cleansing, transforming, and enriching unstructured data to prepare it for LLM pre-training, fine-tuning, and RAG applications.

  • Dispatch
    🤗 PEFT library release enables parameter-efficient fine-tuning of billion‑scale models

    Hugging Face announced the 🤗 PEFT library, which lets users fine‑tune billion‑parameter models on low‑resource hardware by training only a tiny fraction of parameters.