PickRaon-1-mini-preview: Krafton Raon-1 mini, 28B parameter open-source LLM

Raon-1 mini, released by Krafton, is an open-source LLM for text generation. It has a total of 28.7 billion parameters and is distributed under the Apache 2.0 license, allowing for commercial use. It can be loaded directly using custom code from the transformers library on Hugging Face.

1 / 2

The model uses a ChatML-based, text-only SFT protocol. It handles various roles such as system, user, assistant, and tool calls, and can be configured to include or exclude the reasoning process. When making tool calls, it structures and passes function names and arguments in XML tag format.

It is compatible with major inference engines such as vLLM, SGLang, and Docker Model Runner. You can build a server and send requests via an OpenAI-compatible API, and it can also be run as a quantized version in local apps like llama.cpp, Ollama, and LM Studio. Quick start guides are provided for RTX 5090 (32GB) and DGX Spark (GB10) environments.

It supports agentic use cases such as Hermes Agent, Pi, and OpenCode. It is suitable not only for text generation tasks but also for handling complex tasks through tool calls. It has gained attention in the community with 1,177 downloads and 189 likes.