pico-lm/pretokenized-paloma-tinsy
收藏Hugging Face2025-04-16 更新2024-12-14 收录
下载链接:
https://hf-mirror.com/datasets/pico-lm/pretokenized-paloma-tinsy
下载链接
链接失效反馈官方服务:
资源简介:
这是一个包含1024个样本的预分词批次,每个样本的序列长度为2048。数据是从Paloma评估基准中随机采样的,且不在训练语料库中。该数据集的目的是从Pico模型中提取模型激活,用于学习动力学研究。
This is a single batch of 1024 examples each of sequence length 2048. The data is randomly sampled from the Paloma evaluation benchmark, and is not found in the training corpus. The point of this data is to extract model activations from the Pico models that can be used for learning dynamics research.
提供机构:
pico-lm



