Collabora announced on April 17, 2026, the integration of BitNet-style ternary Large Language Model (LLM) inference into ExecuTorch. This implementation leverages ExecuTorch's Vulkan backend to enable efficient AI processing. Presented at PyTorch Conference Europe 2026, this development aims to facilitate the use of much smaller and more bandwidth-efficient LLM models. The primary benefit is portable GPU execution of these models on edge devices, extending advanced AI capabilities to hardware environments with limited resources.