smcleish/scaling-laws-cache
收藏Hugging Face2025-02-07 更新2025-02-15 收录
下载链接:
https://hf-mirror.com/datasets/smcleish/scaling-laws-cache
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是关于宝石模型缩放规律的缓存,包含了delta=1e-4的最小值数据以及delta=1e-3的最小值数据。数据集包含多个特征,如深度、宽度、令牌数、每个令牌的浮点运算数(FLOPs)、总FLOPs、参数数量、带有嵌入的参数数量、6N的FLOPs、预测损失参数、权重衰减比例等。训练集包含13个示例。
This dataset is a cache for the scaling-laws related to gemstone models, including mins for delta=1e-4 and mins for delta=1e-3. The dataset contains multiple features such as depth, width, number of tokens, FLOPs per token, total FLOPs, number of parameters, parameters with embeds, FLOPs_6N, predictive loss parameters, weight decay ratio, etc. The training set contains 13 examples.
提供机构:
smcleish



