tinyllamas

tinyllamas

Provides a series of small language models in the Llama 2 design, trained on the TinyStories dataset and made for Karpathy's llama2.c project. For developers who want small models to try.

Made by karpathy on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds a series of small language models by Andrej Karpathy. They use the Llama 2 design and learned from the TinyStories dataset. The card says they are meant for the llama2.c project.

Who it suits: It suits developers, students and researchers who want a small model to try, read about or build on, rather than a large one.

Industry | Who uses it | How they use it | Example
Software | Developers | Try a small Llama 2 style model in the llama2.c project | Test a model end to end on your own computer
Education | Students | Study how a language model works at small scale | Read through a small model's files
Research | ML researchers | Experiment with training on the TinyStories dataset | Compare the base models with fine-tuned versions
Hobby | Tinkerers | Run a tiny model for fun | Point llama2.c at the files you downloaded

If you need a general chat assistant, you can skip it, since it learned from TinyStories and is meant for the llama2.c project.

Where to use it:

• The llama2.c project, which the card names as the intended use
• Windows, macOS or Linux, where you download the files with Python

Use case | Who | What they get back
Try llama2.c | Developers | Model files made for that project
Study a small model | Students | A small Llama 2 style model to read about
Compare versions | Researchers | 3 quantized and 7 fine-tuned versions made by other people

How to set it up: You need Python with pip, and the llama2.c project to run the files. The card gives no file sizes, so check the file list on the page before you download.

• Python and pip
• The llama2.c project, linked from the card
• Disk space for the files you choose

1. Install the Hugging Face client with pip.
2. Download the repository with the hf command.
3. Follow the llama2.c project's own steps to load a model.

pip install -U huggingface_hub
hf download karpathy/tinyllamas

How to use it: The card shows no run commands. It says only that the models are meant for the llama2.c project, so follow that project's instructions to load one.

Example: You download the files with the hf command, then point llama2.c at one of them. The card gives no run steps, so you follow that project's page from there.

For newcomers and experts: Think of it as a pocket-sized language model made to be simple to study and quick to try. The first small thing to try is the download command above.

The card says only that this is a series of models with the Llama 2 architecture, trained on the TinyStories dataset. It lists no sizes, scores or settings, so check the file list and the llama2.c page for details.

TinyStories dataset
Llama 2 architecture models trained on it
Series published on Hugging Face
Used in the llama2.c project

Good to know: Other versions: Hugging Face lists versions made by other people.

Title: Versions made by other people
Quantized: 3
Fine-tuned: 7

• Popularity: the page has 212 likes, and no downloads were recorded in the last 30 days.
• Short card: the card is a single sentence and gives no sizes, scores or run steps.
• License: MIT, as the License line says.
• Cost to run: no API key, account or paid service is named.
• Data it reaches: the prompts you give your runner, and Hugging Face for the download.
• Not for: the card names no forbidden uses.

On Hugging Face, 13 November 2026

• Downloads in the last 30 days: 0
• Likes: 212
• License: mit
• Quantized versions: 3
• Fine-tunes: 7

Links

• Model: https://huggingface.co/karpathy/tinyllamas
• Files: https://huggingface.co/karpathy/tinyllamas/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download karpathy/tinyllamas

Then follow the usage in its model card. Nothing in this kit runs by itself.