Introduction¶
DeepSeek Harness (DSH) adopts an “everything is a plugin” architecture, allowing users to extend its functionality with plugins. For Partial Wave Analysis (PWA) in particle physics, manually handling physical gates (such as kinematic thresholds, decay-vertex J^P, and C conservation) as well as the iterative fitting workflow is extremely cumbersome. The auto-pwa plugin aims to solve this pain point through AI-driven automation, providing end-to-end support across configuration editing, fitting execution, and iterative convergence.
Core Features¶
This plugin provides the following core capabilities:
* PDG-2026 resonance table lookup: Supports querying resonance names, J^P, J^PC, mass ranges, and decay final states.
* Physical gate configuration editing: Strictly validates configurations; physical gates such as kinematic thresholds, decay-vertex J^P, and C conservation are never exempted.
* ctpwa fitting execution: Integrates the ctpwa fitting engine and supports background task submission.
* Numerical fitting evaluation: Provides numerical diagnostics such as chi2/ndf, pull regions, and partial-wave fractions.
* Goal-driven iterative convergence: Supports goal-based iterative logic, with automatic evaluation and convergence determination.
* SLURM cluster job submission: Supports submitting jobs via a SLURM cluster in environments without a local GPU.
* Partial Wave Analysis toolkit: Provides the auto_pwa_* series of tools, covering lookup, diagnostics, editing, execution, and other stages.
Installation and Enablement¶
You can install and enable it globally via npm.
npm install -g @deepseek-ai/dsh auto-pwa
After installation, the plugin automatically registers the auto-pwa-analysis skill, so there is no need to manually copy it into the ~/.dsh/skills directory.
Typical Usage¶
- Launch the interactive interface (starts the Web GUI by default):
auto-pwa
- Upgrade the plugin:
auto-pwa --upgrade
- Run the analysis in headless mode:
auto-pwa -p headless "完成此文件夹分波"
- Mount a local patch (used during development or when the plugin is unpublished):
pnpm dsh web --patch /absolute/path/to/auto-pwa/patch/auto-pwa.cordis.yml
Notes¶
- Permissions and risks: The plugin runs with the permissions of the current DSH process. Before installing, make sure to review the source code and license.
- Physical gates: Physical gates are never exempted, including kinematic thresholds, decay-vertex J^P, and C conservation.
- Dependency environment: Depends on the YAML runtime.
- Default fitting program: Uses the bundled
scripts/aifit.pyfor fitting by default.
Ending¶
This plugin is maintained by BHXiang, categorized as model inference, and released under the MIT license.