Ollama reports model is 100% on CPU when actually running on GPU

With a Meteor Lake 165U and Windows 11, I'm seeing via Task Manager that in fact the GPU runs inference (GPU Compute very high average utilization for GPU 0 Intel Graphics). But `ollama ps` always incorrectly reports "100% **CPU**" in the "PROCESSOR" column. This is true for all models of any size or quantization that I have tried. I would expect "100% **GPU**" to be reported in these circumstances.