Scaling laws predict the outcome of large, expensive training runs by extrapolating trends from smaller, cheaper runs. On downstream accuracy evaluations, th…
机构:Stanford
来源:arXiv 2611.02270 | AI4Papers 论文推荐平台