MyPracticalTools

English | 简体中文

Small, reusable tools and reference assets for long-running ML, data, and automation workflows.

Included projects

Project	Runtime	Purpose
Practical Run Dashboard	Python 3.10+	Read-only terminal dashboard for checkpointed training and evaluation jobs, including progress, ETA, GPU use, and recoverable state.
watchdogDownloader	Bash on Linux	Resumable manifest-based downloads with low-speed detection, process supervision, and size or SHA-256 verification.
ACA small v0.2	Zstandard JSONL	Versioned train/test dataset package with a manifest, blind-gate metadata, and audit reports.
Dataset design notes	Markdown	Reproducibility, split isolation, replay, validator, reference, and release requirements for the ACA data package.
User workstyle skill	Markdown	A Codex skill for evidence-driven coding, research, and technical communication.

Quick start

Practical Run Dashboard

Install directly from this repository:

python -m pip install \
  "git+https://github.com/Whning0513/MyPracticalTools.git#subdirectory=dashboard"

Copy and edit the example configuration, then start the live terminal view:

practical-dashboard --config dashboard/examples/aca-rl-matrix.json

See the dashboard documentation for JSON output, checkpoint semantics, GPU ownership, and expected status files.

watchdogDownloader

Install the script into a directory on PATH:

git clone https://github.com/Whning0513/MyPracticalTools.git
install -Dm755 MyPracticalTools/watchdogDownloader/wdd "$HOME/.local/bin/wdd"
wdd --help

Create a tab-separated manifest, initialize a download project, and start it:

wdd init /srv/download-state /srv/files ./manifest.tsv
wdd resume /srv/download-state
wdd tui /srv/download-state

See the watchdogDownloader documentation for the manifest format, retry behavior, verification, and tuning controls.

Dataset package

datasets/ACA_small_v0.2/ contains compressed train/test problems and submissions. Use manifest.json as the versioned source of truth and inspect audit_report.json before consuming the package. The accompanying design document distinguishes frozen benchmark evidence from replay output, constructed references, validators, and blind probes.

Development

Run the Python tests:

python -m pip install -e './dashboard[test]'
python -m pytest dashboard/tests -q

Check the Bash script and its interruption/resume integration tests:

bash -n watchdogDownloader/wdd
watchdogDownloader/wdd --version
python -m unittest discover -s watchdogDownloader/tests -v

CI runs both checks for pull requests and changes to main.

License

The original software components, including Practical Run Dashboard, watchdogDownloader, and the user-workstyle skill, are licensed under the MIT License. The dataset package, data-design materials, and any third-party content are excluded unless a separate license is identified. See LICENSE_SCOPE.md for the exact path-level scope.

Contributing

Focused bug reports and pull requests are welcome. Include a minimal reproduction, explain the behavioral change, and state the validation command. Do not include credentials, private datasets, generated caches, or machine- specific absolute paths.