Looped architectures scale quality while keeping parameter count fixed by repeatedly passing activations through the same block of layers. During inference, …
机构:UCSD
来源:arXiv 2610.31633 | AI4Papers 论文推荐平台