Open Source

Command-line tools for the daily work of ML

A small suite of dependency-light command-line tools that I build and maintain to learn the daily work of ML engineering. Each does one thing, runs offline, and fits into CI. Issues and pull requests are welcome.

0 downloads on PyPI in the last month

View the organization on GitHub

Data

dsdiffA git-style diff between two datasets, with distribution drift.0 downloads/monthMore splitcheckDetect rows leaking between train, validation and test splits.0 downloads/monthMore pii-sweepScan datasets for PII before sharing them.0 downloads/monthMore framescanProfile a CSV or Parquet file and write the result as one HTML page.More

Environment and reproducibility

mlenvSnapshot the ML stack to a file and diff two snapshots.0 downloads/monthMore repro-manifestA portable receipt per run to explain why two runs differ.0 downloads/monthMore gpu-gateWait for a free GPU, claim it, and run your command.0 downloads/monthMore

Evaluation

evalgateDecide whether a metric change is a real regression or noise.0 downloads/monthMore slicemapFind the slices where a new model regressed.0 downloads/monthMore pytest-metricguardA pytest fixture that fails the test when an ML metric regresses.More

Serving and cost

servectlServe a model over HTTP in one command, with metrics.0 downloads/monthMore tokenmeterCount tokens and estimate prompt cost before sending.0 downloads/monthMore model-budget-actionFail CI when a model weight or a bundle grows past its budget.More

Training

chainruleA scalar autograd engine for actually learning backpropagation.More expbookA local, serverless experiment tracker: one run is one directory of JSON.More param-lensCount PyTorch model parameters from the editor, without running the code.More

Want to talk data or ML?: hola@jmwebsoluciones.com