On-device inference requires kernels that make effective use of local accelerators. Efficient training and evaluation of models that generate Apple Metal ker…
机构:上交
来源:arXiv 2611.17294 | AI4Papers 论文推荐平台