Unsloth adds native ARM64 Windows CUDA support on NVIDIA RTX Spark

Unsloth says its tools now run natively on ARM64 Windows with CUDA on NVIDIA RTX Spark laptops, for running and training diffusion, multimodal, language and decision models.

Unsloth says training on the platform is 2.7 times faster, a figure the company reports and that the reviewed reports have not independently measured. Users can install through winget or from unsloth.ai/download/windows.

In a separate post, Unsloth says RTX Spark lets users run and train models above 120 billion parameters locally on a laptop, with up to 128GB of unified memory. Its chart compares RTX Spark N1X with a premium SoC and shows 2.7x for fine-tuning with Unsloth.