sachinn1/xl-durel2

XL-DURel-2

This model is an extended variant of XL-DURel.
It has been trained using the Ordinal WiC dataset (training and test splits), as introduced in the paper:
XL-DURel: Finetuning Sentence Transformers for Ordinal Word-in-Context Classification.

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('{MODEL_NAME}')
embeddings = model.encode(sentences)
print(embeddings)

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 10049 with parameters:

{'batch_size': 32, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.AnglELoss.AnglELoss with parameters:

{'scale': 20.0, 'similarity_fct': 'pairwise_angle_sim'}

Parameters of the fit()-Method:

{
    "epochs": 10,
    "evaluation_steps": 2512,
    "evaluator": "WordTransformer.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 1e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 10049,
    "weight_decay": 0.0
}

Full Model Architecture

WordTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

Citing & Authors

@misc{yadav2025xldurelfinetuningsentencetransformers,
      title={XL-DURel: Finetuning Sentence Transformers for Ordinal Word-in-Context Classification}, 
      author={Sachin Yadav and Dominik Schlechtweg},
      year={2025},
      eprint={2507.14578},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2507.14578}, 
}