CITATION.cff

cff-version: 1.2.0
message: "If you use Yaz, please cite it as below."
title: "Yaz: an editable, auditable tiny knowledge model with routing-confidence abstention"
abstract: >-
  Yaz is a sub-1M-parameter, byte-level language model that stores each fact in its own
  addressable decoder column (an "atom"), routes prompts to facts by meaning via a frozen
  sentence embedding, supports create/read/update/delete edits with provable per-edit
  locality, and abstains when its routing confidence is low. A research prototype and an
  honest step in the model-editing lineage (ROME/MEMIT, GRACE, SERAC, PENME) — not a moat.
authors:
  - name: "Tilelli LAB"
    email: hello@tilelli.tech
    city: Marrakech
version: 0.0.1
date-released: 2026-06-17
license: MIT
keywords:
  - knowledge editing
  - model editing
  - selective prediction
  - abstention
  - byte-level language model
  - interpretability
repository-code: "https://github.com/TilelliLab/yaz"
url: "https://github.com/TilelliLab/yaz"