README / README.md
kasparpl's picture
organization card
ad18514 verified
|
Raw History Blame Contribute Delete
689 Bytes
metadata
title: README
emoji: 🎯
colorFrom: gray
colorTo: gray
sdk: static
pinned: false

DOPEST is a software studio from Poland. We build web and mobile products for clients and publish our own open models.

DOPEST 1.0 is our typed-decision model: given a state (text or JSON) and a question with a closed set of options, it returns a calibrated probability for every option in one forward pass. It is a LoRA adapter on Gemma 4 31B with a bidirectional readout on the global-attention layers, trained on rule-engine-generated and verified synthetic decisions, released under Apache 2.0 and evaluated on JevBench.

Code and serving: https://github.com/DOPEST-studio/dopest-model