JallyAI/Nomi-2-Mini

<p align="center"> <img src="https://cdn-uploads.huggingface.co/production/uploads/6921fa6332f7fb129563d495/aR36SrpWzksbcGbcp84pE.png" width="128"> </p>

Nomi 2.0 Mini

Introduction

Introducing Nomi 2 Mini, it was fine tuned on the same data as Nomi 2 and has a very short and efficient reasoning thanks to the RASV reasoning style. Nomi 2 Mini has only 2B parameters, half the parameters of the normal Nomi 2.

If you want to know more about Nomi 2 or RASV, checkout the Nomi 2 model card https://huggingface.com/JallyAI/Nomi-2

🌟 Key Features & Improvements

• —Architecture: Qwen-3.5-2B (requires just ~1.5 GB VRAM).
• —Multilingual Support: Can understand and generate text English and many other languages.
• —Efficiency: Get 100+ tokens/s on consumer hardware, like an RTX 4060. You can use Nomi 2 Mini with an context window of almost 200k tokens

🧠 Training Details

• —Base Model: Qwen/Qwen3.5-2B
• —Fine-tuning: SFT (Supervised Fine-Tuning).
• —Training Tool: Unsloth (for 4-bit optimized training).

😎 Cool License

Feel free to use or improve Nomi! Benchmark results are always welcome.