Tether Launches Cross-Platform LLM Fine-Tuning Framework

Tether, the issuer of the world's largest stablecoin, announced on March 17, 2026, the launch of the first cross-platform framework for fine-tuning large language models on everyday consumer hardware, including smartphones and non-NVIDIA graphics processors. The system, integrated into the company's QVAC Fabric platform, leverages Microsoft's BitNet 1-bit architecture and LoRA low-rank adaptation techniques to enable on-device training of models with up to 13 billion parameters, marking a step toward broader accessibility in AI development.

The framework addresses key barriers in AI model customization by drastically reducing memory and computational demands. BitNet models use binary weights, cutting virtual random access memory usage by up to 77.8 percent compared to 16-bit counterparts like Gemma 3 1B during both inference and fine-tuning. This efficiency allows tasks previously confined to data centers or high-end GPUs to run on standard laptops, mid-range consumer graphics cards, and mobile devices equipped with chips from AMD, Intel, Apple Silicon, Qualcomm Adreno, Arm Mali, or Apple Bionic.

In demonstrations, engineers fine-tuned a 1-billion-parameter BitNet model on a...