Raon-1 nano (GGUF)

Raon-1 nano (GGUF)

Packages the Raon-1 nano language model as GGUF files for llama.cpp, so developers can run a Korean and English chat model on their own computer. Four quant types are listed.

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 GGUF files of Raon-1 nano, a compact KRAFTON AI language model, for use with llama.cpp. The files come in four quant types, listed further down. They were converted from the BF16 weights with a preview branch of KRAFTON's own llama.cpp.

Who it suits: It suits developers who want to run a small Korean and English model locally, on a laptop, a desktop or an Apple silicon Mac, without a Python training stack.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Software development | Developers | Run the model locally through llama.cpp | Testing a chat feature on a laptop with the smallest file
Mobile and edge apps | App teams | Prototype on-device chat in Korean and English | Trying the model before choosing a final format
AI hobbyists | Local model users | Chat with a small bilingual model offline | Asking questions in Korean on a home computer

People who want the original weights for transformers can skip it and use the main Raon-1 nano repository.

Where to use it:

• llama.cpp: the card says the files are for it, on Apple silicon, NVIDIA, AMD and CPU
• Windows, macOS or Linux: download the files with Python
• Other formats: the card also mentions a 4-bit MLX model and a LiteRT-LM bundle, outside this repository

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 machine
Tool-aware assistants | Developers | A model made to follow instructions and call tools
Long documents | Researchers | A context window of 262,144 tokens

How to set it up: You need these first:

• Python on Windows, macOS or Linux
• A llama.cpp build that can read these files, with the card pointing to KRAFTON's preview branch
• Storage for about 1.6 GB at 4-bit, as an estimate

1. Install the Hugging Face hub package.
2. Download the repository files.
3. Load a GGUF file in llama.cpp, following the card.

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

The four file types differ in how tightly the weights are packed.

Title: File types in this repository
Type | What it means
Q4_K_M | 4-bit, the smallest of the four
Q6_K | 6-bit, between the small and large files
Q8_0 | 8-bit, closer to the original quality
F16 | 16-bit floating point, the largest file

How to use it: Download the files, load one in llama.cpp and send prompts in Korean or English. This command fetches them.

hf download KRAFTON/Raon-1-nano-gguf

The card recommends different sampling settings by task.

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 Q4_K_M file, 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, start with the smallest file, Q4_K_M, and ask it one plain question. The card also points to an online playground if you want to try Raon without setting anything up.

If you work in the field, the model is a dense hybrid with 18 gated delta-rule linear attention layers and 6 gated MLA layers, a 256-dimensional embedding per layer and a 262,144-token window. Larger quants cost more memory for fewer rounding losses. The generation config sets the end and padding token IDs but no sampling values.

Download a GGUF file
Load it in llama.cpp
Set temperature and presence penalty
Send a prompt, read the reply

Good to know: The files were made with a preview branch, so check which llama.cpp build the card names before running them.

• License: apache-2.0
• Cost to run: none named
• Data it reaches: only the prompts you give it, with no outside service named

On Hugging Face, 13 November 2026

• Task: text generation
• Parameters: 3.1B
• Weights: about 1.6 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 1,893
• Likes: 22
• License: apache-2.0
• Based on: KRAFTON/Raon-1-nano
• Tags: gguf, raon, llama.cpp, text-generation, conversational

Links

• Model: https://huggingface.co/KRAFTON/Raon-1-nano-gguf
• Files: https://huggingface.co/KRAFTON/Raon-1-nano-gguf/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-gguf

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