neurips2026lrtransfer/LR-Transfer-Trajectory
收藏官方服务：
资源简介：
该数据集记录了12层GPT风格仅解码器Transformer训练过程中的每一步训练和验证指标。每个训练运行存储为一个单独的CSV文件，每行对应一个记录步骤，多列包含JSON编码的每个参数测量值。数据集设计用于支持事后分析，包括损失曲线（训练/验证）、吞吐量和步骤延迟、每层/每参数动态（梯度、优化器输出、参数更新、权重，以及优化器输出与权重之间的对齐）。
This dataset captures per-step training and validation metrics from training runs of a 12-layer GPT-style decoder-only transformer. Each run is stored as a single .csv file in which every row corresponds to one logged step, and several columns hold per-parameter measurements encoded as JSON. The dataset is designed to support post-hoc analysis of: loss curves (train / val), throughput and step latency, per-layer / per-parameter dynamics: gradients, optimizer outputs, parameter updates, weights, and the alignment between optimizer outputs and weights.
