Dataset · Text classification
arabic-decision-benchmark
by NOMEDA nomeda-lab/arabic-decision-benchmark
A unified Arabic typed-decision evaluation benchmark, built to measure a model as a decision engine (choose / score / yes-no over supplied options) rather than a free-text generator.
Dataset Card
A unified Arabic typed-decision evaluation benchmark, built to measure a model as a decision engine (choose / score / yes-no over supplied options) rather than a free-text generator. It merges two complementary evaluation layers: - Core — held-out, genuinely Arabic states from licensed Arabic sources, with benchmark-only (pool-B) question wording. - OpenJev — the repaired new-in-V1 domains (mailroom, amountextraction, phoneextraction, irdecision, entityalignment, silentfailure, sponsorsegment, contextretention, emailselection, citationcontrol, browserdrone) recovered from the existing Arabic OpenJev variants. choice = pick one candidate; score = ordered level; noul = yes/no. Every candidate…
Excerpt from the card by NOMEDA, licensed other.
Details
- Repository
- nomeda-lab/arabic-decision-benchmark
- Publisher
- NOMEDA
- Task category
- Text classification
- Tags
- arabic, typed-decisions, benchmark
- Size category
- 1K<n<10K
- Languages
- ar, en
- Revision
- 198e8e9d19f089881f53c3e3d52b85d398a65c85
- Last updated
- 2026-10-08
Files
5 files, 30.2 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| arabic_decision_benchmark.jsonl | Data | 14.0 MB | 0409b6b14f84 |
| arabic_decision_benchmark_core.jsonl | Data | 5.0 MB | — |
| arabic_decision_benchmark_primary.jsonl | Data | 11.2 MB | e42f56754d06 |
| README.md | Documentation | 4.4 KB | — |
| .gitattributes | Repository | 2.6 KB | — |
License and Download
- License
- other
- Access
- No access gate
Released by NOMEDA through its official repository on Hugging Face.