Raon-1 nano (MLX)

Raon-1 nano (MLX)

Provides Raon-1 nano as 4-bit MLX weights for Apple silicon, so developers can run a Korean and English chat model locally on a Mac. For app builders and local model users.

Made by KRAFTON 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 the MLX weights of Raon-1 nano, a compact KRAFTON AI language model, made for Apple silicon. The weights use 4-bit GPTQ quantization with a group size of 64. The card points to an MLX section for a BF16 export too.

Who it suits: It suits developers and enthusiasts who want a small Korean and English chat model running on a Mac with Apple silicon.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Software development | Mac developers | Run the model locally with MLX | Testing a chat feature on an Apple silicon laptop
Korean-language products | Product teams | Prototype Korean conversation features on a Mac | Trying prompts before choosing a deployment format
AI hobbyists | Local model users | Chat with a small bilingual model offline | Asking questions in Korean or English on a personal Mac

People on Windows or Linux hardware without Apple silicon can skip it and use the main Raon-1 nano repository or the GGUF files.

Where to use it:

• MLX on Apple silicon: the weights are made for it
• Windows, macOS or Linux: you can download the files with Python, though the weights target Apple silicon
• Other runtimes: the card lists Transformers, vLLM, llama.cpp and LiteRT-LM for Raon-1 nano, each with its own files

Title: Jobs it does
Use case | Who | What they get back
Local chat | Individuals | Replies in Korean or English from a model on their own Mac
Instruction following | Developers | Text written to match the instruction given
Tool calling | App builders | A choice of tool inside a conversation

How to set it up: You need these first:

• Python on Windows, macOS or Linux, to download the files
• A Mac with Apple silicon and MLX, to run the weights

1. Install the Hugging Face hub package.
2. Download the repository files.
3. Load the weights with MLX, following the model card.

pip install -U huggingface_hub
hf download KRAFTON/Raon-1-nano-mlx

How to use it: Download the weights, load them with MLX and send prompts in Korean or English. This command fetches the files.

hf download KRAFTON/Raon-1-nano-mlx

The card lists sampling settings by task for Raon-1 nano.

Title: Recommended sampling settings
Task | Temperature | Presence penalty
General chat, knowledge questions, instruction following | 0.1 | 1.5
Agentic work or code | 1.0 | Not set

Thinking is on by default in the chat template. The output token limit counts both thinking and the final answer, so raise it for longer replies.

Example: You load the 4-bit weights on a Mac, set temperature to 0.1 and presence penalty to 1.5, and ask a general question in Korean. A text reply comes back, and a short cut-off means the token limit needs raising.

For newcomers and experts: If you are new to this, think of it as a small chat model packed for Apple laptops. Download it and try one short question once MLX is installed.

If you work in the field, the base model is a hybrid of linear and full attention layers with a per-layer embedding and a 262,144-token window. This repository carries the 4-bit GPTQ export in MLX form. The generation config sets the end and padding token IDs but no sampling values.

Download the MLX weights
Load them with MLX on Apple silicon
Set temperature and presence penalty
Send a prompt, read the reply

Good to know: These weights are 4-bit, and the card points to its MLX section for a BF16 export.

• License: apache-2.0
• Cost to run: none named
• Data it reaches: only the prompts you give it, with no outside service named
• Not for: hardware without Apple silicon, since the weights are made for it

On Hugging Face, 13 November 2026

• Task: text generation
• Library: mlx
• Parameters: 3.1B
• Weights: about 12.5 GB in U32
• Formats: Safetensors
• Downloads in the last 30 days: 94
• Likes: 16
• License: apache-2.0
• Based on: KRAFTON/Raon-1-nano
• Tags: safetensors, raon1_nano, raon, text-generation, conversational

Links

• Model: https://huggingface.co/KRAFTON/Raon-1-nano-mlx
• Files: https://huggingface.co/KRAFTON/Raon-1-nano-mlx/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download KRAFTON/Raon-1-nano-mlx

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