Tether releases the cross-platform BitNet LoRA framework, supporting large model training and inference on consumer-grade GPUs and smartphones.

According to Mars Finance, Tether CEO Paolo Ardoino disclosed that the Tether AI team has released a new version of QVAC Fabric, integrating the cross-platform BitNet LoRA framework, enabling training and inference of billion-parameter models on consumer-grade GPUs and smartphones. The new QVAC Fabric LLM is the first to achieve cross-platform operation of BitNet LoRA fine-tuning and inference on AMD, Intel, Apple Metal, and mobile GPUs. On flagship devices, GPU inference speed is 2 to 11 times faster than CPU, and memory usage is reduced by up to 90% compared to full-precision models. The Tether team has already fine-tuned models with up to 3.8 billion parameters on flagship phones such as the Pixel 9, S25, and iPhone 16, and achieved fine-tuning of models with up to 13 billion parameters on the iPhone 16. The relevant code has been open-sourced on GitHub.