| --- |
| language: he |
| license: apache-2.0 |
| base_model: onlplab/alephbert-base |
| pipeline_tag: text-classification |
| tags: |
| - natural-language-inference |
| - hebrew |
| - hebnli |
| - HeRE |
| --- |
| |
| # AlephBERT fine-tuned on HebNLI |
|
|
| `onlplab/alephbert-base` fine-tuned as a 3-class NLI classifier (entailment / neutral / contradiction) on [HebNLI](https://github.com/NNLP-IL/HebNLI). It is one of the Strategy 1 denoisers of the HeRE paper: the premise is a Hebrew Wikipedia passage and the hypothesis is a verbalised Wikidata triple; the entailment probability decides whether the passage expresses the relation. |
|
|
| ## Training |
|
|
| AdamW, learning rate 2e-5, batch size 64, linear warm-up over 10% of steps, weight decay 0.01, maximum sequence length 250, 8,000 steps, evaluation every 200 steps; the checkpoint with the best validation macro-F1 was retained (step 4,600). Single run, seed 42. |
|
|
| ## Results |
|
|
| | | HebNLI dev macro-F1 | HebNLI test macro-F1 | HeRE gold validation F1 | |
| |---|---|---|---| |
| | AlephBERT | 0.763 | 0.781 | 0.618 (Strategy 1, LLM-generated hypothesis) | |
|
|
| ## Labels |
|
|
| `id2label`: `{"0": "讛讬住拽", "1": "住转讬专讛", "2": "谞讬讟专诇讬"}` |
|
|
| ## Usage |
|
|
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| tok = AutoTokenizer.from_pretrained("ronke21/hebnli-alephbert") |
| model = AutoModelForSequenceClassification.from_pretrained("ronke21/hebnli-alephbert") |
| enc = tok(premise, hypothesis, return_tensors="pt", truncation=True, max_length=250) |
| probs = model(**enc).logits.softmax(-1) |
| ``` |
|
|
| ## License |
|
|
| The fine-tuned weights are released under the base model's license (apache-2.0). Paper resources: dataset https://huggingface.co/datasets/ronke21/HeRE, code https://github.com/Ronke21/HeRE. |
|
|
| ## Citation |
|
|
| Keinan, R., Cohen, A. D. N., and Tsarfaty, R. (2026). HeRE: A Novel Benchmark for Hebrew Relation Extraction via LLM-Guided Denoising of Knowledge Graph Alignments. In *Proceedings of AACL 2026*. |
|
|
