On March 17, stablecoin issuer Tether announced that its AI platform QVAC Fabric has launched the world's first cross-platform LoRA fine-tuning framework for Microsoft BitNet (1-bit LLM), enabling billion-parameter language models to be trained and inferred on common hardware, including laptops, consumer GPUs and smartphones.
The official statement indicates that the framework significantly lowers the threshold for GPU memory and computing power required for AI model training, and supports Intel, AMD, Apple Silicon, and various mobile GPUs (such as Adreno, Mali, and Apple Bionic).
In testing, a BitNet model with approximately 125 million parameters could be fine-tuned in about 10 minutes on a Samsung S25; a model with 1 billion parameters took about 1 hour and 18 minutes on a Samsung S25 and about 1 hour and 45 minutes on an iPhone 16. The team even successfully fine-tuned a model with 13 billion parameters on an iPhone 16.
In terms of performance, BitNet models can achieve inference speeds of 2 to 11 times faster on mobile GPUs compared to CPUs. Meanwhile, tests show that BitNet-1B can reduce GPU memory usage by up to 77.8% compared to 16-bit models in inference and fine-tuning tasks.
Paolo Ardoino stated that this technology aims to reduce reliance on large-scale cloud computing and dedicated AI hardware, enabling AI model training to be completed on local devices and laying the foundation for new paradigms such as decentralized AI and federated learning.
