0f7c460f1b
Keep existing yaml entries unchanged; enable SciCode sandbox in run.py and host-side image build without GDPval. Co-authored-by: Cursor <cursoragent@cursor.com>
40 lines
1.6 KiB
Bash
Executable File
40 lines
1.6 KiB
Bash
Executable File
#!/bin/bash
|
||
# 在宿主机本地构建 SciCode 沙箱镜像（不从 Docker Hub 拉 scicode-benchmark）。
|
||
# 用法与 Terminal Bench 预加载相同：先在宿主机准备好镜像，再 docker run 评测容器并挂 docker.sock。
|
||
#
|
||
# bash bash/images_load/preload_k3_sandbox_images.sh
|
||
#
|
||
# 注意：
|
||
# - 评测仍在 evalscope 容器里跑，沙箱是宿主机 Docker 再起的子容器
|
||
# - SciCode 首次运行还会按内容哈希打 tag，有本地层缓存就不会再拉 Hub
|
||
set -euo pipefail
|
||
|
||
PROJECT_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
|
||
SCICODE_DIR="$PROJECT_ROOT/evalscope/evalscope/benchmarks/scicode/docker"
|
||
H5_URL="https://modelscope.cn/datasets/evalscope/SciCode/resolve/master/test_data.h5"
|
||
|
||
echo "==> 检查基座镜像（已有则跳过 pull）"
|
||
if docker image inspect python:3.11-slim >/dev/null 2>&1; then
|
||
echo " ok python:3.11-slim"
|
||
else
|
||
echo " docker pull python:3.11-slim"
|
||
docker pull python:3.11-slim
|
||
fi
|
||
|
||
echo "==> SciCode: 确保 test_data.h5"
|
||
if [ ! -f "$SCICODE_DIR/test_data.h5" ]; then
|
||
echo " downloading test_data.h5 ..."
|
||
curl -L --fail -o "$SCICODE_DIR/test_data.h5" "$H5_URL"
|
||
fi
|
||
ls -lh "$SCICODE_DIR/test_data.h5"
|
||
|
||
echo "==> docker build scicode-benchmark:latest"
|
||
docker build -t scicode-benchmark:latest "$SCICODE_DIR"
|
||
|
||
echo ""
|
||
echo "完成。本地镜像："
|
||
docker images --format 'table {{.Repository}}\t{{.Tag}}\t{{.Size}}' | grep -E 'REPOSITORY|scicode-benchmark|python'
|
||
echo ""
|
||
echo "下一步：docker run 评测镜像，挂 -v /var/run/docker.sock:/var/run/docker.sock"
|
||
echo "SciCode 需要 run.py 打开 sandbox（已加入 SANDBOX_DATASETS）。"
|