Skip to content

v1.7.0 — Domain Presets, Halt Evidence, Pluggable Scorer

Choose a tag to compare

View all tags
@anulum anulum released this 01 Mar 23:51
· 3006 commits to main since this release
v1.7.0
a5b9e55

What's New

  • Domain presets: DirectorConfig.from_profile("medical"|"finance"|"legal"|"creative"|"customer_support") with tuned thresholds, NLI/reranker flags, and logic/fact weights
  • Structured halt evidence: stream_tokens(scorer=...) populates HaltEvidence with top-K contradicting chunks, NLI scores, and suggested action
  • Pluggable scorer backend: CoherenceScorer(scorer_backend="deberta"|"onnx"|"minicheck", onnx_path=...)
  • Batched MiniCheck: native list API for batch NLI scoring
  • False-halt rate: 0.0% on 20 good passages (heuristic mode), asserted in CI
  • Competitor one-pager: summary comparison table (Director-AI vs NeMo vs Lynx vs GuardrailsAI vs SelfCheckGPT)
  • Repo hygiene: dependabot, ROADMAP, updated SECURITY versions, copyright headers, all GH Actions bumped to latest

Install

pip install director-ai==1.7.0

Full Changelog

https://github.com/anulum/director-ai/blob/main/CHANGELOG.md

Assets 4
Loading