README.md

metadata

title: Meridian-Research
emoji: 🧭
colorFrom: gray
colorTo: red
sdk: static
pinned: false

Meridian

Pretraining small base models from scratch, experimenting with the experimental.

Compact language models, released with training details — architecture,
data sources, token budgets, benchmarks, and limitations.

Novely-33m-chat
_{32.5M · SFT-aligned, GGUF}

Cagliari-114m
_{114M · Base, 8.16B tokens}

Zedev-134m
_{134.47M · Base, 18.1B tokens}

Gast-Mini
_{7.07M · Base, custom BPE}

Contact: drago.py on Discord