Skip to content

Latest commit

 

History

History
26 lines (21 loc) · 1.08 KB

File metadata and controls

26 lines (21 loc) · 1.08 KB

Phase 4 Functioning Model — 2026-04-14

Cognition-core now includes a real native functioning model path.

What was added

  • adapters/exemplar_adapter.py
  • scripts/training/train_exemplar_candidate.py
  • scripts/execution/run_functioning_candidate.py
  • tests/test_functioning_model.py
  • docs/FUNCTIONING_MODEL_STATUS_2026-04-14.md

What it does

The new exemplar adapter loads a deterministic local artifact built from curated JSONL corpora already in the repo and the supporting package ingestions.

Verified result

Candidate: darpa_functional

  • artifact built successfully
  • benchmark run completed successfully
  • scored output file generated successfully
  • current overall weighted score: 0.3458
  • full candidate gate still passes for the scaffold lane

Honest status

This is a real runnable model path and closes the biggest honesty gap in the repo. It is not yet the final GGUF-trained candidate path. The remaining hard blockers are still the true external trainer / merge / GGUF exporter loop and retention validation against a produced quantized artifact.