Tether Launches Cross-Platform BitNet LoRA Framework for Training Billion-Parameter Models on Consumer Devices

PANews, March 21: According to an official announcement, Tether has launched a cross-platform BitNet LoRA fine-tuning framework within QVAC Fabric, optimizing the training and inference of Microsoft BitNet (1-bit LLM). This framework significantly reduces computational and memory requirements, enabling training and fine-tuning of billion-parameter models on laptops, consumer-grade GPUs, and smartphones. This solution marks the first time BitNet models can be fine-tuned on mobile GPUs, including Adreno, Mali, and Apple Bionic. Testing shows that a 125M parameter model can be fine-tuned in approximately 10 minutes, a 1B model in about one hour, and the framework even scales to 13B parameter models on mobile devices. Additionally, the framework supports heterogeneous hardware such as Intel, AMD, and Apple Silicon, and is the first to enable 1-bit LLM LoRA fine-tuning on non-NVIDIA devices. In terms of performance, BitNet model inference on mobile GPUs is 2 to 11 times faster than on CPUs, while reducing VRAM usage by up to 77.8% compared to traditional 16-bit models. Tether states that this technology has the potential to break reliance on high-end computational power and cloud infrastructure, advancing AI training toward decentralization and localization, and providing a foundation for emerging applications such as federated learning.