Improve ollama's Output Speed

I have noticed that ollama always outputs content at a fixed speed, and most of the time, the GPU is not fully utilized(0% load), indicating that the bottleneck in generating content is not in the GPU. At the same time, I have observed that during content output, one CPU core consistently has a high load(70% load), suggesting that the bottleneck might be in the CPU decoding process. Therefore, could you consider implementing multi-threaded decoding to leverage the multi-core capabilities of the CPU to accelerate the output speed? I have already searched Google for discussions on this issue but did not find anything relevant. If my analysis is incorrect or the issue cannot be resolved, you can directly close this issue. Thank you.