Training Lens · 训练视窗 v0.1.0

Training Lens · 训练视窗的首个公开版本：在 DeepSeek Harness 中，用已有 TensorBoard 标量日志查看训练进展。

• 任意主要 / 辅助指标、可选改善方向、原始值与 EMA 曲线。
• 中英文界面、运行切换、自动刷新和窄屏布局。
• training_lens 工具支持启动、状态与关闭；只管理插件自己的监控进程。
• 日志留在本机；安装包不包含模型、训练数据或个人配置。

安装 / Install

已发布 npm：dsh-training-lens@0.1.0。推荐通过 DSH 安装：

dsh plugin --profile desktop add dsh-training-lens@0.1.0

npm 公开下载包的 SHA256 与已验证的 GitHub Release 安装包完全一致；已通过独立 DSH profile 从 npm 安装验证。

也可使用 GitHub Release 安装包：

dsh plugin --profile desktop add https://github.com/GU143312/dsh-training-lens/releases/download/v0.1.0/dsh-training-lens-0.1.0.tgz

安装后重载插件或重启 DSH，再提供日志目录及训练环境 Python 的完整路径。也可以下载 .tgz 后将本地路径传给同一条 add 命令。详见 中文说明。

Requires DSH 0.2.1-rc.2, Node 22+, and an existing Python environment with TensorBoard. The plugin does not install Python dependencies automatically.

验证 / Validation

Windows, Linux and macOS CI passed: five Node tests and ten Python tests per platform, syntax checks and package validation. The published tarball was downloaded, checksum-verified, installed into an isolated profile, and booted in the actual DSH 0.2.1-rc.2 host. Its registered tool started the dashboard, returned synthetic scalar events over HTTP 200, and stopped it successfully. Browser checks covered Chinese/English and responsive layouts. No model API call or real training was required.

For a single downloadable handoff file, Training-Lens-v0.1.0.zip contains the reviewed source, the installable .tgz, and a short Chinese installation guide. Install the .tgz from inside it, rather than passing the ZIP directly to DSH.

This is an independent community plugin available through npm and GitHub Releases. It is not an official DSH endorsement. MIT license. Demonstration screenshots use synthetic data.