Turkish MMLU Leaderboard

Turkish MMLU Leaderboard

Ranks AI models on a Turkish version of the MMLU benchmark, with filters, browsable model answers, per-section charts and a form for submitting new models. For people who compare models on Turkish.

Made by alibayram 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: This web app ranks AI models by how they score on a Turkish MMLU test. It also lets you read each model's answers, see charts by subject section, and submit a new model for evaluation. The whole interface is in Turkish.

Who it suits: It suits researchers and engineers who need to compare language models on Turkish tasks. Model authors can also use the submission form to get a model checked and queued.

Industry | Who uses it | How they use it | Example
Research | NLP researchers | Compare models on Turkish language tasks | Filter the ranking to one model family
Software | ML engineers | Pick a model that handles Turkish well | Look at section charts for the weak spots
Model authors | Open model builders | Register a model for evaluation | Check a model against the Ollama library first
Education | Students | Learn how a benchmark leaderboard works | Browse questions next to each model's answers

Anyone who does not work with Turkish language models can skip it.

Where to use it: The app runs in three places.

• A web browser, on the page of the Space at Hugging Face
• Windows, macOS or Linux, as a local copy run with Python
• Docker, from the Dockerfile in the repository

Use case | Who | What they get back
Ranking | Researchers | A model ranking you can filter by family and quantization level
Model answers | Analysts | Questions with each model's answers, searchable and paged
Section analytics | Engineers | Charts and statistical summaries for each subject section
Model submission | Model authors | A compatibility check, a RAM estimate and a spot in the evaluation queue

How to set it up: The online page needs nothing installed. A local copy needs the following.

• Python 3.8 or newer
• pip
• The packages listed in requirements.txt
• Docker, only if you prefer a container

1. Install Python.
2. Install the Hugging Face hub package and download a copy of the Space.
3. Move into the new folder and install Gradio and the packages in requirements.txt.
4. Start the app with python app.py.
5. For Docker, build the image and run it with port 7860 published.

pip install -U huggingface_hub
hf download alibayram/turkish_mmlu_leaderboard --repo-type space --local-dir turkish_mmlu_leaderboard
cd turkish_mmlu_leaderboard
pip install gradio
pip install -r requirements.txt
python app.py

docker build -t turkish-mmlu-leaderboard .
docker run -p 7860:7860 turkish-mmlu-leaderboard

How to use it: Open the page and move between its sections. Filter the ranking, browse the questions and answers, read the charts, or fill in the submission form. On a local copy, this command starts the page.

python app.py

Tool | What it does
Leaderboard | Ranks models and filters them by family and quantization level
Model answers | Lets you search and page through questions and each model's answers
Performance analytics | Draws interactive charts per section and compares models
Model submission | Validates a model in the Ollama library and estimates the RAM it needs

Example: You filter the ranking to one model family and one quantization level, then open the section charts. You see how the models that remain compare in each subject, and you can read their answers question by question.

For newcomers and experts: If you are new to benchmarks, a leaderboard is a sorted list of scores on one shared test. Open the page, look at the top of the ranking first, then open the answers section to see what a score is made of.

If you know the field, the app reads three Hugging Face datasets, one each for the ranking, the model answers and the section results. It keeps a cache and refreshes the data every 30 minutes, and it falls back to stored data when a refresh fails. The submission check parses Ollama pages with beautifulsoup4 and requests, then estimates RAM at 16-bit, 8-bit and 4-bit and applies a 20GB limit. The code splits into a data manager, a model validator and one module per section of the interface.

Three datasets feed the app
A cache refreshes every 30 minutes
Sections show ranking, answers, charts
Submission checks a model on Ollama

Good to know: The interface, messages and table labels are all in Turkish. Scores update on their own, so you do not need to reload by hand.

• License: cc-by-nc-4, a Creative Commons non-commercial license
• Cost to run: none named
• Data it reaches: three Hugging Face datasets, and the Ollama library when you validate a submitted model
• Limit: the compatibility check applies a 20GB RAM ceiling
• Not for: commercial use, since the license is non-commercial

On Hugging Face, 13 November 2026

• Likes: 30
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard
• Files: https://huggingface.co/spaces/alibayram/turkish_mmlu_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.