EnterpriseRAG Bench Leaderboard

EnterpriseRAG Bench Leaderboard

Ranks retrieval-augmented generation systems on EnterpriseRAG-Bench, a benchmark built on company knowledge questions, with tabs to browse questions and compare systems side by side. For teams choosing a RAG system.

Made by onyx-dot-app on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: EnterpriseRAG Bench Leaderboard is a Gradio app that ranks RAG systems on EnterpriseRAG-Bench, a benchmark about a company's internal knowledge. The base set holds 500 questions, and each system is scored on its answers. You can open it online, or build and run your own copy.

Who it suits: It suits teams who are comparing retrieval-augmented generation systems for use on internal documents, and people who maintain the benchmark.

Industry | Who uses it | How they use it | Example
Enterprise software | Platform teams | Compare RAG systems before choosing one | Open the ranked table and read the order
AI research | Researchers | Study how systems do on different question types | Open the per-system breakdown
Quality assurance | Evaluators | Check individual answers against gold answers | Step through one system's questions
Benchmark maintenance | Maintainers | Add a new system to the board | Follow the steps for a new submission

Skip it if you want to run your own evaluation from scratch, since the app displays results that were scored elsewhere.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Docker on your own Windows, macOS or Linux computer
• Python on your own computer, for a direct local run

Use case | Who | What they get back
Rank systems | Teams | A table of systems in order of score
Inspect one system | Evaluators | Its answers question by question
Compare two systems | Researchers | Their answers side by side

How to set it up: The online page needs nothing. A local copy needs Docker, or Python with the packages in requirements.txt, and Python with pip to fetch the files.

• Docker, or Python with pip
• Python packages from requirements.txt, for a direct run
• Windows, macOS or Linux

1. Install the Hugging Face hub library.
2. Download the Space's files into a folder named EnterpriseRAG-Bench-Leaderboard.
3. Move into that folder.
4. Build the Docker image.
5. Run the container with port 7860 published, then open that port in your browser.

pip install -U huggingface_hub
hf download onyx-dot-app/EnterpriseRAG-Bench-Leaderboard --repo-type space --local-dir EnterpriseRAG-Bench-Leaderboard
cd EnterpriseRAG-Bench-Leaderboard
docker build -t enterpriserag-bench-leaderboard .
docker run -it -p 7860:7860 enterpriserag-bench-leaderboard

To skip Docker, install the packages and start the app directly. It serves on port 7860.

pip install -r requirements.txt
python app.py

How to use it: Open the page and move between the tabs.

Tool | What it does
Leaderboard tab | The main table of systems, ranked
Submission overview tab | A breakdown for each system by question type
Data viewer tab | Lets you browse the questions, the gold answers and the facts
Submission viewer tab | Shows one system's answers question by question
Side-by-side tab | Puts two systems next to each other

To add a system, put its answer, result and corrected question files in data/raw_data, add a display name in systems.yaml, then regenerate the display data.

your_key: "Your System Display Name"

python transform_raw_data.py

Example: You open the leaderboard tab to see how the systems rank. You then switch to the side-by-side tab and compare how two of them answered the same question.

For newcomers and experts: Start with the leaderboard tab for the overall order, then open the data viewer to read a few questions and their gold answers.

Each system's files share a short key, written in lowercase with underscores. Keys already registered include bm25, langchain, ragflow, vertexai and weaviate_verba. The transform script finds systems from the result files and checks that every required file and mapping exists. A new system should go live only when all 500 questions have valid, populated records, and a partial run should not be pushed. The files under final_display_data are generated and are not meant to be edited by hand.

Raw files go in data/raw_data
A display name is added to systems.yaml
transform_raw_data.py builds the display data
The Gradio app shows the tabs

Good to know:

• Scale: The base question set holds 500 questions.
• Rule: Do not publish a partial run. Every record needs an answer and filled scores.
• License: mit.
• Cost to run: none named.
• Data it reaches: The evaluation files bundled in the repository. The record names one dataset, EnterpriseRAG-Bench.

On Hugging Face, 13 November 2026

• Likes: 40
• Built with: docker
• License: mit

Links

• Space: https://huggingface.co/spaces/onyx-dot-app/EnterpriseRAG-Bench-Leaderboard
• Files: https://huggingface.co/spaces/onyx-dot-app/EnterpriseRAG-Bench-Leaderboard/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.