EttinX Cross-Encoder: Natural Language Inference (NLI)

This cross encoder performs sequence classification for contradiction/neutral/entailment labels. This has drop-in compatibility with comparable sentence transformers cross encoders.

To train this model, I added teacher logits to the all-nli dataset dleemiller/all-nli-distill from the dleemiller/ModernCE-large-nli model. This significantly improves performance above standard training.

This 68m architecture is based on ModernBERT and is an excellent candidate for lightweight CPU inference.

Features

High performing: Achieves 87.98% and 88.67% (Micro F1) on MNLI mismatched and SNLI test.
Efficient architecture: Based on the Ettin-68m encoder design (68M parameters), offering faster inference speeds.
Extended context length: Processes sequences up to 8192 tokens, great for LLM output evals.

Performance

Model	MNLI Mismatched	SNLI Test	Context Length	# Parameters
dleemiller/ModernCE-large-nli	0.9202	0.9110	8192	395M
dleemiller/ModernCE-base-nli	0.9034	0.9025	8192	149M
cross-encoder/nli-deberta-v3-large	0.9049	0.9220	512	435M
cross-encoder/nli-deberta-v3-base	0.9004	0.9234	512	184M
dleemiller/EttinX-nli-s	0.8798	0.8967	8192	68M
cross-encoder/nli-distilroberta-base	0.8398	0.8838	512	82M
dleemiller/EttinX-nli-xs	0.8380	0.8820	8192	32M
dleemiller/EttinX-nli-xxs	0.8047	0.8695	8192	17M

Extended NLI Evaluation Results

F1-Micro scores (equivalent to accuracy) for each dataset.

Model	finecat	mnli	mnli_mismatched	snli	anli_r1	anli_r2	anli_r3	wanli	lingnli
dleemiller/finecat-nli-l	0.8152	0.9088	0.9217	0.9259	0.7400	0.5230	0.5150	0.7424	0.8689
tasksource/ModernBERT-large-nli	0.7959	0.8983	0.9229	0.9188	0.7260	0.5110	0.4925	0.6978	0.8504
dleemiller/ModernCE-large-nli	0.7811	0.9088	0.9205	0.9273	0.6630	0.4860	0.4408	0.6576	0.8566
tasksource/ModernBERT-base-nli	0.7595	0.8685	0.8979	0.8915	0.6300	0.4820	0.4192	0.6632	0.8118
dleemiller/ModernCE-base-nli	0.7533	0.8923	0.9035	0.9187	0.5240	0.3950	0.3333	0.6464	0.8282
dleemiller/EttinX-nli-s	0.7251	0.8765	0.8798	0.9128	0.3360	0.2790	0.3083	0.6234	0.8012
dleemiller/EttinX-nli-xs	0.7013	0.8376	0.8380	0.8979	0.2780	0.2840	0.2800	0.5838	0.7521
dleemiller/EttinX-nli-xxs	0.6842	0.7988	0.8047	0.8851	0.2590	0.3060	0.2992	0.5426	0.7018

Usage

To use EttinX for NLI tasks, you can load the model with the Hugging Face sentence-transformers library:

from sentence_transformers import CrossEncoder

# Load EttinX model
model = CrossEncoder("dleemiller/EttinX-nli-s")

scores = model.predict([
    ('A man is eating pizza', 'A man eats something'),
    ('A black race car starts up in front of a crowd of people.', 'A man is driving down a lonely road.')
])

# Convert scores to labels
label_mapping = ['contradiction', 'entailment', 'neutral']
labels = [label_mapping[score_max] for score_max in scores.argmax(axis=1)]
# ['entailment', 'contradiction']

Training Details

Pretraining

We initialize the `` weights.

Details:

Batch size: 256
Learning rate: 1e-4
Attention Dropout: attention dropout 0.1

Fine-Tuning

Fine-tuning was performed on the dleemiller/all-nli-distill dataset.

Validation Results

The model achieved the following test set micro f1 performance after fine-tuning:

MNLI Unmatched: 0.8798
SNLI: 0.8967

Model Card

Architecture: Ettin-encoder-68m
Fine-Tuning Data: dleemiller/all-nli-distill

Thank You

Thanks to the Johns Hopkins team for providing the ModernBERT models, and the Sentence Transformers team for their leadership in transformer encoder models.

Citation

If you use this model in your research, please cite:

@misc{moderncenli2025,
  author = {Miller, D. Lee},
  title = {EttinX NLI: An NLI cross encoder model},
  year = {2025},
  publisher = {Hugging Face Hub},
  url = {https://huggingface.co/dleemiller/EttinX-nli-xxs},
}