Hugging Face's logo

oddadmix
/
Nawah-RuleCheck-v2

Text Classification
Transformers
Safetensors
Arabic
llama
arabic
rule-checking
compliance
moderation
tiny-model
on-device
text-embeddings-inference
Model card Files
xet
 Community
Nawah-RuleCheck-v2

Commit History

  1. ship LADDER.json for reproducibility

    oddadmix
    Verified signature
    3f050ad

  2. Card regenerated from eval_all.py - one harness, all four rungs

    oddadmix
    Verified signature
    7b130c9

  3. Replace with LR-swept checkpoint (1e-4/10ep): unseen 0.9799 -> 0.9980

    oddadmix
    Verified signature
    e937509

  4. ship LADDER.json for reproducibility

    oddadmix
    Verified signature
    fbb825c

  5. ship train_rules_cls.py for reproducibility

    oddadmix
    Verified signature
    dc69a70

  6. ship ood_cases.json for reproducibility

    oddadmix
    Verified signature
    898e78f

  7. ship cf_cases.json for reproducibility

    oddadmix
    Verified signature
    f3a9d4b

  8. ship eval_all.py for reproducibility

    oddadmix
    Verified signature
    f2d43bf

  9. Card regenerated from eval_all.py

    oddadmix
    Verified signature
    0d8868f

  10. counterfactual_eval: select model via CF_MODEL

    oddadmix
    Verified signature
    a91f64c

  11. Model card: minimal-pair result re-measured on v2 (0.998, was v1's 1.000), size/latency table, out-of-pool wording limitation

    oddadmix
    Verified signature
    3d00f66

  1. Nawah-RuleCheck-v2: 51.8M Arabic rule checker, 0.980 on unseen rule wording⁠

    oddadmix
    Verified signature
    489a82b

  2. initial commit

    oddadmix
    Verified signature
    068c6aa