Tether releases QVAC cross-platform BitNet LoRA framework: supports training AI models with billions of parameters on consumer-grade devices

Odaily reported that, according to an official announcement, Tether has launched a cross-platform BitNet LoRA fine-tuning framework within QVAC Fabric, enabling training and inference optimization for Microsoft BitNet (1-bit LLM). This framework significantly reduces computing power and memory requirements, allowing billion-parameter models to be trained and fine-tuned on laptops, consumer-grade GPUs, and smartphones.

This solution achieves, for the first time, BitNet model fine-tuning on mobile GPUs (including Adreno, Mali, and Apple Bionic). Tests show that a 125M parameter model can be fine-tuned in about 10 minutes, while a 1B model can be completed in roughly 1 hour, and even on mobile devices, it can scale up to a 13B parameter model.

Additionally, the framework supports heterogeneous hardware such as Intel, AMD, and Apple Silicon, and for the first time enables 1-bit LLM LoRA fine-tuning on non-NVIDIA devices. In terms of performance, BitNet models achieve inference speeds on mobile GPUs that are 2 to 11 times faster than CPUs, while memory usage is reduced by up to 77.8% compared to traditional 16-bit models.

Tether stated that this technology is expected to break the dependence on high-end computing power and cloud infrastructure, promote the decentralization and localization of AI training, and provide a foundation for new application scenarios such as federated learning.