Skip to content

refactor(api): align model and session vocabulary with lab roles - #195

Merged
KartikP merged 2 commits into
unified-model-interface-v2from
refactor/lab-vocabulary
Oct 22, 2026
Merged

KartikP merged 2 commits into
unified-model-interface-v2from
refactor/lab-vocabulary

Conversation

@KartikP

@KartikP KartikP commented Oct 22, 2026

Copy link
Copy Markdown
Contributor

Summary

Use lab-oriented configuration names: extractor, answer, policy, and intervention. Introduce BufferedSession and respond() while retaining existing parameter names, imports, and emit() implementations. Consolidate overlapping session helpers without changing trial resets or scoring behavior.

Compatibility tests cover inherited response methods, conflicting settings, four input modalities, policy/behavior output equivalence, and intervention cleanup.

Validation

  • Coordinated CPU profile: core 809, vision 239, language 70, unified 1,310 passed; 43 allowed skips, no unexpected skips.
  • Broader local suites: core 843 and unified 1,182 passed (overlapping counts).
  • 15 offline notebooks executed in fresh kernels.
  • Data-dependent full-suite attempts were limited by S3 access, model downloads, and environment requirements; scientific/GPU qualification was not rerun.

Companion unified PR updates integrations, guides, notebooks, and the core pin. Merge core first. No package publication or deployment.

@KartikP
KartikP merged commit 023f0bd into unified-model-interface-v2 Oct 22, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant