| cff-version: 1.2.0 | |
| title: "Tilelli — a small routed byte-LM with verifiable claims" | |
| message: "If you use this kit, please cite it as below." | |
| version: "0.1.0" | |
| date-released: "2026-05-24" | |
| authors: | |
| - name: "Tilelli LLM Team" | |
| license: Apache-2.0 | |
| repository-code: "https://github.com/TilelliLab/Tilelli-llm" | |
| abstract: > | |
| A 10 M-parameter byte-level language model with a 3-pathway heterogeneous | |
| block. Trained on a single GPU, runs on a laptop CPU. Every numerical | |
| claim in the README is bound to a reproduce script that exits non-zero | |
| if the bundled checkpoint fails to produce the documented number. | |
| Ships verified positive results (held-out IDK gate, NEO false-inability | |
| rate) alongside verified negative results (router-entropy is not free | |
| metacognition at this scale; abstain heads do not transfer modularly; | |
| the router cannot be retrained on subset distributions without breaking | |
| generation). | |
| keywords: | |
| - small language model | |
| - mixture of experts | |
| - routing | |
| - calibration | |
| - negative results | |
| - reproducibility | |