Topic: row split

The ggml/llama.cpp project released a fix for idle threads in specific multiplication kernels used for neural networks with fewer than 1024 elements. This update generalizes a previous row split to additional related kernels, improving thread utilization and potentially enhancing performance.