Changelog

v0.1.0 — 2026-07-25

Initial public evidence release.

Hardware-aware search across quantization, pruning, compilation, batching,
and runtime configurations.
PyTorch, CUDA, ONNX Runtime, TensorRT, and Ray execution paths.
Pareto-frontier and constraint-aware recommendation logic.
Local FastAPI inference-bundle export.
Verified Apple Silicon CPU and NVIDIA Tesla T4 evidence across three real
pretrained models and public labeled evaluation data.
Reproducibility records, failed-trial retention, cost proof, and termination
proof.

This release does not claim memory reduction, external users, or deployment.