multiclinner-enigma-ro-procedure-xlm-roberta-large-full

MultiClinNER ENIGMA RO Procedure XLM-RoBERTa-large Model

This is a Romanian clinical Named Entity Recognition (NER) model for the Procedure entity type, fine-tuned from xlm-roberta-large on the Romanian portion of the MultiClinNER shared task. The model was developed by Team Enigma at the Faculty of Mathematics and Informatics, Sofia University.

Description

The model is based on xlm-roberta-large and fine-tuned for token classification to identify PROCEDURE clinical entities.

This version was trained on the combined train + development sets.

Training Hyperparameters

The following hyperparameters were used during training:

• Maximum Sequence Length: 512
• Stride: 128
• Learning Rate: 2e-5
• Training Epochs: 5
• Per Device Train Batch Size: 4
• Gradient Accumulation Steps: 2 (Effective Batch Size = 8)
• Weight Decay: 0.01
• Seed: 42
• Precision: bf16=True, tf32=True

Post-processing

The official submission pipeline includes an additional rule-based post-processing step applied after model inference. The post-processing step handles span cleanup, duplicate removal, overlapping span resolution, punctuation trimming, and restoration of entity spans using the original document offsets.

The post-processing code is available in the official repository.

Usage

Note: The snippet below provides a quick out-of-the-box pipeline demonstration for raw text inference using standard Hugging Face subtoken aggregation. For official evaluation and submission, please utilize the dedicated word-alignment and offset restoration script from the repository.

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

repo_id = "MonikaPeteva/multiclinner-enigma-ro-procedure-xlm-roberta-large-full"

tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForTokenClassification.from_pretrained(repo_id)

ner = pipeline(
    "token-classification",
    model=model,
    tokenizer=tokenizer,
    aggregation_strategy="simple"
)

text = "Patient was diagnosed with pneumonia."
predictions = ner(text)

print(predictions)

Intended Use

This model is intended for clinical NER research in the context of the MultiClinNER task.

Limitations

The model should be used with awareness that the official submission results may depend on the additional post-processing pipeline.

Repository

• Official Repository: