Skip to content

MPSBoost 0.2.0 Release Audit

0.2.0 was the first formal MPS histogram engine release. It established the real compiled Apple Silicon backend, deterministic data/binning path, regression GBDT, model save/load, cache safety, and release verification workflow.

Included:

  • real compiled MPS/Metal backend foundation;
  • native device diagnostics and GPU smoke kernel;
  • deterministic quantization and compact binned representation;
  • real regression GBDT training and prediction;
  • model save/load;
  • S6 benchmark evidence for both MPS wins and small-data regressions;
  • cache diagnostics, explicit cache creation, and safe cache clearing;
  • wheel, license, dynamic-link, and PyPI verification.

Not included:

  • classifiers;
  • forest/ExtraTrees/CatBoost-like families;
  • advanced objectives;
  • anomaly/ranking estimators;
  • large real-world dataset matrix.