Tether launched the first cross-platform LoRA fine-tuning framework for Microsoft’s BitNet models, through its QVAC Fabric platform, with the goal of enabling the training and inference of language models of up to one billion parameters on consumer hardware, including laptops, conventional GPUs and modern smartphones.
This framework eliminates the dependency on enterprise-grade NVIDIA infrastructure or cloud services, which until now represented a nearly insurmountable access barrier for organizations outside the top technology tier.
Benchmarks published by the company show that the BitNet-1B model consumes up to 77.8% less VRAM than Gemma-3-1B in 16-bit, and that inference on mobile GPUs is between two and eleven times faster than on CPU. On a Samsung S25, fine-tuning a 125-million-parameter model on a biomedical dataset of approximately 300 documents completes in ten minutes. A 1B model requires one hour and eighteen minutes on the same device, while the iPhone 16 allows scaling up to 13B-parameter models.
Paolo Ardoino, CEO of Tether, stated that when model training depends on centralized infrastructure, innovation stagnates and the balance of the ecosystem becomes fragile. He also noted that the company will continue allocating resources to ensure that artificial intelligence is accessible locally on any device.
Disclaimer: Crypto Economy Flash News are based on verified public and official sources. Their purpose is to provide fast, factual updates about relevant events in the crypto and blockchain ecosystem.
This information does not constitute financial advice or investment recommendation. Readers are encouraged to verify all details through official project channels before making any related decisions.
