Public evaluate pipeline and report schema v1
The initial public release on GitHub and PyPI: core evaluate pipeline, guard chain, schema v1, and the first docs/CLI surface.
Release: InvarLock 0.2.0 - First public GitHub and PyPI release
Highlights
- Core compare/evaluate pipeline + guard chain.
- Report schema v1 + usable CLI entry points.
- Torch-optional install with extras + initial docs.
This is the starting line: the first time the whole idea is available publicly as a cohesive tool, not just a repo of experiments. 0.2.0 ships the core evaluate loop and guard chain, plus the report schema and CLI surface that makes it usable without living in the codebase.
It also sets the tone for versioning expectations: until 1.0.0, minor releases can include breaking changes. That’s not a warning so much as a “this is moving fast” label—so upgrades should be deliberate, not accidental.
If you’re new to InvarLock, this is the release that gives you the baseline mental model for everything that follows.
For the immutable release record, read the tagged CHANGELOG.md for v0.2.0.
More in Release
Continue through nearby posts in the same reading thread.
Release
Quantization-aware adapters and safe device movement
First-class quantization metadata, safer device movement across quantized models, auto-routing based on checkpoint info, and major test coverage expansion.
Release
Large-model reload stability and B200 controls
Fixes a GPU memory leak during reload fallback, hardens B200 scripts, and adds practical controls for acceptance ranges and overhead measurement.
Release
Calibration, determinism, and regression protection
`invarlock calibrate` arrives, determinism utilities mature, and regression harness + golden tracking help prevent silent policy drift.