preflight-ml

Version updated for https://github.com/Rusheel86/preflight to version v0.1.1.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

The preflight GitHub Action is a pre-training validation tool for PyTorch pipelines designed to detect common issues like NaNs, label leakage, incorrect channel ordering, and dead gradients before model training begins. By running a series of automated checks in seconds, it helps prevent wasted computational resources and ensures data and model integrity prior to initiating training. This action integrates seamlessly into CI pipelines and fails builds automatically if critical issues are detected.

Release notes

Initial release of preflight — pre-flight checks for PyTorch pipelines.

What’s included

  • 10 checks across fatal/warn/info severity tiers
  • NaN/Inf detection, label leakage, shape mismatch, gradient check
  • Normalisation sanity, channel ordering, VRAM estimation, class imbalance
  • Split sizes, duplicate sample detection
  • .preflight.toml config file support
  • JSON output mode for CI integration

Usage

```yaml

Install via pip

``` pip install preflight-ml ```