{"id":1779,"date":"2026-08-19T13:07:43","date_gmt":"2026-08-19T13:07:43","guid":{"rendered":"https:\/\/blog.openzeka.com\/en\/?p=1779"},"modified":"2026-08-19T13:07:43","modified_gmt":"2026-08-19T13:07:43","slug":"local-model-fine-tuning-with-unsloth-on-dgx-spark","status":"publish","type":"post","link":"https:\/\/blog.openzeka.com\/en\/local-model-fine-tuning-with-unsloth-on-dgx-spark\/","title":{"rendered":"Local Model Fine-Tuning with Unsloth on DGX Spark"},"content":{"rendered":"

In this tutorial, you will fine-tune a large language model on a single DGX Spark using Unsloth.<\/p>\n

The Spark can be used as a standalone computer by connecting a monitor and keyboard, or as a server accessed remotely from another computer. In this tutorial, we will connect to the Spark remotely, install the necessary software, and fine-tune the model.<\/p>\n

We will use GPT-OSS 20B (a 21-billion parameter Mixture-of-Experts model) as the model, Unsloth as the fine-tuning framework, and Reinforcement Learning (RL) as the training method. Unsloth will load the model with 4-bit quantization (QLoRA) and train only small LoRA adapters, achieving up to 70% memory savings. The training will teach the language model to play the game “2048”.<\/p>\n

This tutorial consists of three parts:<\/p>\n

    \n
  • Setup:<\/strong> Building the Docker image and starting the container<\/li>\n
  • Training:<\/strong> Running the notebook, loading the model, and GRPO training<\/li>\n
  • Results:<\/strong> Testing the fine-tuned model and shutdown<\/li>\n<\/ul>\n

    This tutorial is for a single Spark; fine-tuning does not require multiple Sparks.<\/p>\n

    \n

    Setup<\/h2>\n

    1. Connecting to the Spark<\/h3>\n

    If you are connecting to the Spark remotely for the first time, you need to find its IP address. Connect a monitor and keyboard to the Spark, log in, and run the following command in the terminal:<\/p>\n

    <\/div>\n<\/div>