ChainThink report, on March 17, stablecoin issuer Tether announced the launch of its AI platform QVAC Fabric, introducing the world's first cross-platform LoRA fine-tuning framework tailored for Microsoft BitNet (1-bit LLM), enabling billion-parameter language models to undergo training and inference on standard hardware—including laptops, consumer-grade GPUs, and smartphones.

Officially, the framework significantly reduces the memory and computational requirements for AI model training, supporting Intel, AMD, Apple Silicon, and various mobile GPUs such as Adreno, Mali, and Apple Bionic.

In testing, a 125 million-parameter BitNet model was fine-tuned in approximately 10 minutes on a Samsung S25; a 1-billion-parameter model completed fine-tuning in about 1 hour 18 minutes on the same device, and in roughly 1 hour 45 minutes on an iPhone 16. The team even successfully fine-tuned a 13-billion-parameter model on the iPhone 16.

Performance-wise, BitNet models achieve inference speeds 2 to 11 times faster on mobile GPUs compared to CPUs. Additionally, test results show that BitNet-1B reduces GPU memory consumption by up to 77.8% during both inference and fine-tuning tasks when compared to 16-bit models.

Paolo Ardoino stated this technology aims to reduce dependency on large-scale cloud computing and specialized AI hardware, enabling on-device AI model training and laying the foundation for novel paradigms such as decentralized AI and federated learning.