Evaluation 文件说明
Evaluation 内容分为两部分保存:
- 镜像大文件:Hugging Face Bucket
KouShi2/ailab-code-evaluation-images - 其余代码、配置和评测数据:Dataset Repo
KouShi2/ailab-code-data的evaluation/目录
Bucket 保留以下原始目录结构,所有 TAR 均保持原文件名,不重新压缩:
tb1/images/
tb2/images/
完整下载与恢复
先安装并登录 Hugging Face CLI:
python3 -m pip install -U hf
hf auth login
下载 Dataset Repo 中的代码和评测数据:
mkdir -p /mnt/shared-storage-gpfs2/intern-agent-dev/shikou/hf
hf download KouShi2/ailab-code-data \
--repo-type dataset \
--include "evaluation/**" \
--local-dir /mnt/shared-storage-gpfs2/intern-agent-dev/shikou/hf
将 Bucket 中的镜像同步回同一个 evaluation 目录:
hf buckets sync \
hf://buckets/KouShi2/ailab-code-evaluation-images \
/mnt/shared-storage-gpfs2/intern-agent-dev/shikou/hf/evaluation
恢复完成后的关键路径为:
/mnt/shared-storage-gpfs2/intern-agent-dev/shikou/hf/evaluation/tb1/images/
/mnt/shared-storage-gpfs2/intern-agent-dev/shikou/hf/evaluation/tb2/images/
如果目录中带有 SHA256SUMS,进入对应目录后执行:
sha256sum -c SHA256SUMS