Note Try the models in the browser: a state, typed questions, and a probability per option.

TigreGotico/jevasio-4b-v2

Text Classification • Updated about 8 hours ago

Note The 4B trained further on the V2 data: 0.839 accuracy on the V2 test split, against 0.743 for the first 4B.

TigreGotico/jevasio-0.8b-v2

Text Classification • Updated about 8 hours ago • 11 • 1

Note The 0.8B trained further on the V2 data: 0.775 accuracy on the V2 test split, against 0.660 for the first 0.8B.

Note The second version: 335k typed decisions, every V1 record plus variety, intents, NLI, classification and generated pt-PT tasks.

TigreGotico/jevasio-4b

Text Classification • Updated about 8 hours ago • 11

Note The larger model: 0.803 accuracy on the pt-PT test split, against 0.729 for stock Kev-4B.

TigreGotico/jevasio-0.8b

Text Classification • Updated about 8 hours ago • 10

Note The small model: 0.721 accuracy on the pt-PT test split, against 0.564 for stock Kev-0.8B.

Note The training and evaluation data: 192k typed decisions in pt-PT, from the amalia-llm datasets.