@marcobarlo
marcobarlo 暂无简介
A heterogeneous hardware acceleration library focused on efficient KV cache transfer operators (H2D/D2H), designed for large model training and inference scenarios.
