Tether Unveils 13B-Parameter Bitnet Framework for Consumer Devices as It Pushes Edge-First AI
Updated
Updated · Computerworld · May 28
Tether Unveils 13B-Parameter Bitnet Framework for Consumer Devices as It Pushes Edge-First AI
1 articles · Updated · Computerworld · May 28
Tether said its new Bitnet LLM fine-tuning framework can run and tune 13-billion-parameter models on consumer hardware, including phones such as the Samsung S25 and iPhone 16 and regular PCs.
The system adds Vulkan and Metal GPU backends to Bitnet, widening support beyond its original Bitnet.cpp engine and enabling cross-platform inference and LoRA fine-tuning on heterogeneous consumer GPUs, including mobile chips.
A dynamic tiling technique is used to work around Vulkan mobile GPU buffer limits, which Tether said lets the framework preserve Bitnet’s low-compute efficiency while operating on widely supported hardware rather than NVIDIA-specific CUDA.
Tether framed the release as part of a broader open-source push through its QVAC SDK, aimed at local-first AI and peer-to-peer delegated inference as companies with under $100 million in revenue still lag larger firms in AI scaling.
Tether promises to democratize AI, but what is its business model and how will the company actually profit from this?
Can running AI on phones truly challenge big tech, or will their massive data advantage always keep them ahead?