Is RAG-FiT open source?

Yes — IntelLabs/RAG-FiT is open source, released under the Apache-2.0 license.

What language is RAG-FiT written in?

IntelLabs/RAG-FiT is primarily written in Python.

How popular is RAG-FiT?

IntelLabs/RAG-FiT has 768 stars on GitHub.

Where can I find RAG-FiT?

IntelLabs/RAG-FiT is on GitHub at https://github.com/IntelLabs/RAG-FiT.

← all repositories

IntelLabs/RAG-FiT

Make your LLM actually read the manual

RAG-FiT generates training datasets from live retrieval pipelines so you can fine-tune LLMs to better exploit external context, then verifies they actually used it.

★768 stars Python RAG · Search LLMOps · Eval Language Models

View on GitHub ↗ Homepage ↗

Not currently ranked — collecting fresh signals.

star history

What it does

RAG-FiT (formerly RAG Foundry) is a modular toolkit that turns RAG pipelines into training material. It ingests raw data, runs retrieval and prompt assembly, persists the results in a model-agnostic format, and feeds them into parameter-efficient fine-tuning via PEFT and TRL. After training, it runs inference and evaluates outputs with metrics that can inspect citations, reasoning traces, and retrieval quality—not just the final text.

The interesting bit

Instead of treating RAG as a fixed retrieval-then-prompt layer, the framework materializes the entire interaction—retrieval results, prompts, metadata—as a reusable dataset. This lets you iterate on the training of the model itself, not just the prompt or the vector index, and evaluate whether the model actually leveraged the retrieved evidence.

Key highlights

Four-stage pipeline: dataset creation, PEFT training, inference, and evaluation, all wired together.
Hydra-driven configuration-as-code, supporting hierarchical overrides and multi-job dispatch to SLURM or Ray.
Evaluation suite includes RAGAS, Deepeval, BERTScore, and custom metrics that can consume retrieval context, citations, and attributions.
Datasets are stored in a consistent, model-independent input-output format with full metadata preserved.

Caveats

Explicitly disclaimed as “not an official Intel product.”

Verdict

Worth a look if you’re hitting a ceiling with prompt engineering and want to train your model to better ingest retrieved knowledge. Skip it if you need a managed vector store or a no-code RAG builder.

Frequently asked

What is IntelLabs/RAG-FiT?: RAG-FiT generates training datasets from live retrieval pipelines so you can fine-tune LLMs to better exploit external context, then verifies they actually used it.
Is RAG-FiT open source?: Yes — IntelLabs/RAG-FiT is open source, released under the Apache-2.0 license.
What language is RAG-FiT written in?: IntelLabs/RAG-FiT is primarily written in Python.
How popular is RAG-FiT?: IntelLabs/RAG-FiT has 768 stars on GitHub.
Where can I find RAG-FiT?: IntelLabs/RAG-FiT is on GitHub at https://github.com/IntelLabs/RAG-FiT.