Back to blog

Fail-closed baseline pairing with lower-memory retries

CI/release baseline pairing is fail-closed (pairing evidence is required), and adapters reduce peak memory during retries via chunked snapshot/restore.

1 min readInvarLock Team
Two rows interlock in short, separately bounded sections, expressing required pairing with work performed in smaller chunks.

Release: InvarLock 0.3.4 - Required pairing evidence and chunked adapter snapshots

Highlights

  • CI/Release baseline pairing now fails closed with strict parity checks.
  • Report generation requires paired-windows evidence (non-perfect pairing is rejected).
  • Chunked snapshot/restore and evidence-pack workflow helpers reduce retry and long-run friction.

This release makes baseline pairing a checked contract instead of a convention. If a CI/release run says it is paired to a baseline, it now has to carry valid evaluation_windows evidence, and the system enforces dataset, tokenizer, and masking parity instead of letting ambiguous comparisons slip through.

Report generation is equally strict: paired_windows evidence is required, and non-perfect pairing is rejected. That gives reviewers a clearer answer to the most important question: when a report says it is paired, it really is paired.

On the practical side, chunked HF adapter snapshot/restore helps reduce peak memory during retries, and the new evidence-pack workflow helpers (run_suite, scheduler/queue utilities, and model creation tooling) make longer evidence runs easier to operate without weakening the pairing rules.

Sources

Website documentation explains the maintained workflow. The changes described here belong to InvarLock v0.3.4; the tagged release record preserves that historical scope.

More in Release

Explore nearby related posts.

Release

Deterministic evidence packs and safer perplexity runs

Evidence packs gain a deterministic bash test suite and better runtime helpers, window selection becomes stable/offline, and perplexity runs get safer around bad token IDs.

Release

Token-weighted paired statistics and stricter release gates

Token-weighted paired bootstrap lands across the pipeline, strictness toggles expand, and CI/release pairing expectations become explicit and enforceable.

Release

Measurement contracts for CI and release verification

Reports now record and enforce estimator measurement contracts under CI/release profiles, and evidence pack suites can cleanly split calibration vs execution.