Unsloth updates Docker image to train and run 500-plus models locally

Unsloth says its Docker image lets users train and run 500-plus models locally through a graphical interface or notebooks. A related announcement lists multi-user accounts for Unsloth Desktop, custom Jupyter Notebook themes and titles, expandable cells, RDNA1 and RDNA2 support, and Windows ARM64 CUDA support. That announcement also claims faster GRPO and 2x faster image diffusion inference with FP8/INT8 support.