otagent-jupiter-2026-05-13
收藏资源简介:
该数据集是一个conda环境快照,旨在精确复现在JSC Jupiter集群(基于NVIDIA GH200 aarch64架构)上为otagent项目配置的计算环境。数据集通过conda-pack工具打包,保留了原始环境的完整依赖关系,特别是本地构建的vLLM 0.16.0 wheel文件及其所有传递依赖。数据集包含以下文件:打包的环境存档文件(otagent-jupiter-2026-05-13.tar.gz)、构建日志(pack.log)、conda环境导出文件(env.yml)以及pip冻结输出(pip_freeze.txt)。源环境使用Python 3.12,并针对CUDA 13和PyTorch 2.9.x栈进行配置。关键软件包包括vLLM 0.16.0、torch 2.9.x和针对aarch64架构预构建的flash-attn。数据集排除了三个以可编辑模式安装的软件包(Harbor、OpenThoughts-Agent及其子模块llamafactory),这些需要在恢复后手动克隆和安装。该环境快照专门为aarch64(Grace)架构和CUDA 13驱动栈构建,不兼容x86_64系统。主要用途是确保研究或开发工作能在与原始环境一致的软件栈中复现,适用于需要特定版本vLLM和PyTorch配置的AI模型服务或实验场景。
This dataset is a conda environment snapshot designed to precisely replicate the computing environment configured for the otagent project on the JSC Jupiter cluster (based on NVIDIA GH200 aarch64 architecture). The dataset is packaged using the conda-pack tool, preserving the complete dependency relationships of the original environment, especially the locally built vLLM 0.16.0 wheel file and all its transitive dependencies. The dataset includes the following files: a packed environment archive file (otagent-jupiter-2026-05-13.tar.gz), build logs (pack.log), a conda environment export file (env.yml), and a pip freeze output (pip_freeze.txt). The source environment uses Python 3.12 and is configured for the CUDA 13 and PyTorch 2.9.x stack. Key software packages include vLLM 0.16.0, torch 2.9.x, and flash-attn pre-built for the aarch64 architecture. The dataset excludes three packages installed in editable mode (Harbor, OpenThoughts-Agent, and its submodule llamafactory), which need to be manually cloned and installed after restoration. This environment snapshot is specifically built for aarch64 (Grace) architecture and CUDA 13 driver stack, and is not compatible with x86_64 systems. The primary purpose is to ensure that research or development work can be reproduced in a software stack consistent with the original environment, suitable for AI model service or experimental scenarios requiring specific versions of vLLM and PyTorch configurations.




