openjev
Datasets
All datasets matching “openjev”Open-Jev
Open-Jev: typed decision datasets
Open-Jev turns a state and a question into a typed decision: a yes/no probability, a distribution over choices, independent label probabilities, or a discrete numeric/ordinal decision. This repository publishes twelve separate, frozen data configs from the Open-Jev project, together with original manifests, exact raw records, source code and reconstruction instructions.
These are controlled, mostly synthetic tasks and reference labels. They are… See the full description on the dataset page: https://huggingface.co/datasets/ZefanCai/Open-Jev.open-jev-laya-benchAPUS-OpenJev-Eval-Frozen80
APUS-OpenJev Eval Frozen80
Private, fixed development evaluation panel: 80 decisions from five task families, 79 parent groups. This is the same panel previously used for official Jev, xDAN-openJet low/high, Laya, and checkpoint comparisons. This publication packages existing data without resampling, changing labels, candidate order, or input content.
This is not an independent test set. It has repeatedly informed debugging and checkpoint selection. Repeated runs measure… See the full description on the dataset page: https://huggingface.co/datasets/apus-ailab/APUS-OpenJev-Eval-Frozen80.OpenJev-Vision-Research-v0.1
OpenJev Vision Research v0.1
12,832 image records, with public provenance, original synthetic scenes,
and programmatically derived decision questions.
This is an experimental research dataset for visual posterior learning and
compositional decisions, released with OpenJev.
It is not a reproduction of TypeSafe's proprietary Jev model or training method.
Three separate configurations
Config
Images
What the labels mean
License
synthetic
8,192
Exact… See the full description on the dataset page: https://huggingface.co/datasets/IamBusy/OpenJev-Vision-Research-v0.1.openjev-mixture
OpenJev training mixture
279 classification and multiple-choice tasks, 323,466 rows, normalised into
one typed-decision format so a single model can be trained across all of
them. Assembled from tasksource plus
two curated ordinal datasets.
Built for OpenJev. The format is
plain JSONL, so nothing here requires that code.
Composition
primitive
tasks
what it is
choice
234
pick one of K options
noul
34
yes/no
score
11
one level on an ordered scale… See the full description on the dataset page: https://huggingface.co/datasets/s1lv3rj1nx/openjev-mixture.openjev-heldout
OpenJev held-out generalization suite
Seven tasks for measuring whether a typed-decision model transfers to
schemas it has never been trained on. Built to answer one question honestly:
does the model generalize, or has it seen this before?
Used by OpenJev, but the format is
plain JSONL and nothing here depends on that code.
The suite
task
primitive
K
rows
chance
clinc_oos
choice
151
600
0.007
banking77
choice
77
600
0.013
massive_intent
choice
60… See the full description on the dataset page: https://huggingface.co/datasets/s1lv3rj1nx/openjev-heldout.