BERT Hash Cross-Encoder: Semantic Similarity (STS)

Cross encoders are high performing encoder models that compare two texts and output a 0-1 score. I've found the cross-encoders/roberta-large-stsb model to be very useful in creating evaluators for LLM outputs. They're simple to use, fast and very accurate.

The BERT hash uses a bucketing technique with projection to decrease the size of the embedding parameters (all <1M parameters). These models are very small and good for inference at the edge.

---

Features

• Performance: Achieves Pearson: 0.7595 and Spearman: 0.7474 on the STS-Benchmark test set.
• Efficient architecture: Based on the BERT Hash model architecture, offering lightweight models.
• Extended context length: Processes sequences up to 8192 tokens, great for LLM output evals.
• Diversified training: Pretrained on dleemiller/wiki-sim and fine-tuned on sentence-transformers/stsb.

---

Performance

---

Usage

To use sts-bert-hash for semantic similarity tasks, you can load the model with the Hugging Face sentence-transformers library:

from sentence_transformers import CrossEncoder

# Load CrossEncoder model
model = CrossEncoder("dleemiller/sts-bert-hash-nano", trust_remote_code=True)

# Predict similarity scores for sentence pairs
sentence_pairs = [
    ("It's a wonderful day outside.", "It's so sunny today!"),
    ("It's a wonderful day outside.", "He drove to work earlier."),
]
scores = model.predict(sentence_pairs)

print(scores)  # Outputs: array([0.9184, 0.0123], dtype=float32)

Output

The model returns similarity scores in the range [0, 1], where higher scores indicate stronger semantic similarity.

---

Training Details

Pretraining

The model was pretrained on the pair-score-sampled subset of the dleemiller/wiki-sim dataset. This dataset provides diverse sentence pairs with semantic similarity scores, helping the model build a robust understanding of relationships between sentences.

• Classifier Dropout: a somewhat large classifier dropout of 0.15, to reduce overreliance on teacher scores.
• Objective: STS-B scores from dleemiller/MocernCE-large-sts.

Fine-Tuning

Fine-tuning was performed on the sentence-transformers/stsb dataset.

Validation Results

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

• Pearson Correlation: 0.7595
• Spearman Correlation: 0.7474

---

Model Card

• Architecture: bert-hash-nano
• Tokenizer: Custom tokenizer trained with modern techniques for long-context handling.
• Pretraining Data: dleemiller/wiki-sim (pair-score-sampled)
• Fine-Tuning Data: sentence-transformers/stsb

---

Thank You

Thanks to the NeuML team for providing the BERT Hash 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{stsnano2025,
  author = {Miller, D. Lee},
  title = {Bert Hash STS: An STS cross encoder model},
  year = {2025},
  publisher = {Hugging Face Hub},
  url = {https://huggingface.co/dleemiller/sts-bert-hash-pico},
}

---

License

This model is licensed under the MIT License.