The llama.cpp project released a new version that can load and run the SenseNova U1 dense understanding branch, including the U1.5‑8B‑MoT model, on a standard CPU. It also adds conversion tools for the model’s vision encoder, allowing image inputs of any resolution. Generation branches are still blocked, so the update only supports understanding tasks. The changes were co‑authored by Sigbjørn Skjæret from Hugging Face.
Why it matters
CPU users can now run the high‑resolution SenseNova U1 vision model without needing a GPU.
