WVY Model Releases
WVY 1.7 1.5M on Kaggle and BbyWVY 360M on Hugging Face are public WVY-family model drops. They are built around the same Starpower direction: efficient reasoning, cleaner training signals, and more useful behavior per parameter.
The releases
| Model | Size | Role |
|---|---|---|
wvy-1.7-1.5m | 1.5M | Public Kaggle release for fast WVY experiments and small-model iteration. |
BbyWVY-360m | 360M | WVY-family model release on Hugging Face for open model access. |
Start low, punch up
We're a new company and we train like one: starting at the bottom and carrying every lesson up the scale. WVY 1.7 1.5M and BbyWVY 360M make that visible. The goal is not the biggest model. The goal is the most useful model per parameter: cleaner datasets, less noise, better reasoning traces, multi-perspective interaction from wvy.world, and expert thinking patterns from the Savvy dataset work.
Training direction
WVY training is aimed at models that can use information to solve problems. Instead of rewarding only immediate answer prediction, we care about reading the available context, organizing it, checking assumptions, and producing a grounded result. That is why the small-model path matters: with a cleaner dataset and sharper training loop, a smaller model can become more reliable for the specific reasoning work we care about.
Using the models
Open the Kaggle release to run and inspect WVY 1.7 1.5M, or open the Hugging Face repo for BbyWVY 360M.
# WVY 1.7 1.5M
# first public WVY small-model drop
# run it on Kaggle
# https://www.kaggle.com/code/starpowertechnology/wvy-1-7-1-5m/edit
# BbyWVY 360M
# WVY-family model release
# https://huggingface.co/StarpowerTechnology/BbyWVY-360m