Skip to content

Gavel base

chukfinleyOpen weightsNot yet run

Von's mechanism trained deliberately for the decision task, small and multilingual, with a fitted temperature and the card's own negative results.

Specifications

Checkpoint
chukfinley/gavel-base
Publisher
chukfinley
Mechanism
NLI entailment head, one hypothesis per option
Backbone
ModernBERT-base, three-way entailment head trained for decisions
Parameters
150M
License
MIT
Runs as
open weights, self-hosted
Library
transformers
Primitives
Choice, Score, Noul
Pricing
none — your compute

Published numbers

Reported by the authors or by third parties, on their own suites. Not comparable to the leaderboard above; attributed so you can check them.

Model cardchukfinley

Gavel base model card

Measured on: Tool selection, CLINC, BoolQ, ANLI, WANLI, Belebele, MMLU, and a TypeSafe business fixture

Tool selection / CLINC / BoolQ
0.993 / 0.947 / 0.785
TypeSafe business fixture
0.409 vs published Jev 0.883
ECE after temperature (T = 2.27)
0.024

Author-reported, with its own negative results listed.

Frequently asked questions

How does Gavel base compare to Jev?

Gavel base has not been run through the benchmark yet; the figures on this page are its authors' own and are not comparable to the leaderboard until a row exists.

Is Gavel base open source?

Yes. chukfinley/gavel-base is released under MIT and runs on your own hardware with the transformers library.

Can Gavel base extract entities?

No. Gavel base answers questions about text and never returns spans; use a GLiNER-family model for extraction.