<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

Model Card: EUBERT

Overview

• Model Name: EUBERT
• Model Version: 1.1
• Date of Release: 16 October 2023
• Model Architecture: BERT (Bidirectional Encoder Representations from Transformers)
• Training Data: Documents registered by the European Publications Office
• Model Use Case: Text Classification, Question Answering, Language Understanding

Model Description

EUBERT is a pretrained BERT uncased model that has been trained on a vast corpus of documents registered by the European Publications Office. These documents span the last 30 years, providing a comprehensive dataset that encompasses a wide range of topics and domains. EUBERT is designed to be a versatile language model that can be fine-tuned for various natural language processing tasks, making it a valuable resource for a variety of applications.

Intended Use

EUBERT serves as a starting point for building more specific natural language understanding models. Its versatility makes it suitable for a wide range of tasks, including but not limited to:

1. Text Classification: EUBERT can be fine-tuned for classifying text documents into different categories, making it useful for applications such as sentiment analysis, topic categorization, and spam detection.

2. Question Answering: By fine-tuning EUBERT on question-answering datasets, it can be used to extract answers from text documents, facilitating tasks like information retrieval and document summarization.

3. Language Understanding: EUBERT can be employed for general language understanding tasks, including named entity recognition, part-of-speech tagging, and text generation.

Performance

The specific performance metrics of EUBERT may vary depending on the downstream task and the quality and quantity of training data used for fine-tuning. Users are encouraged to fine-tune the model on their specific task and evaluate its performance accordingly.

Considerations

• Data Privacy and Compliance: Users should ensure that the use of EUBERT complies with all relevant data privacy and compliance regulations, especially when working with sensitive or personally identifiable information.

• Fine-Tuning: The effectiveness of EUBERT on a given task depends on the quality and quantity of the training data, as well as the fine-tuning process. Careful experimentation and evaluation are essential to achieve optimal results.

• Bias and Fairness: Users should be aware of potential biases in the training data and take appropriate measures to mitigate bias when fine-tuning EUBERT for specific tasks.

Conclusion

EUBERT is a pretrained BERT model that leverages a substantial corpus of documents from the European Publications Office. It offers a versatile foundation for developing natural language processing solutions across a wide range of applications, enabling researchers and developers to create custom models for text classification, question answering, and language understanding tasks. Users are encouraged to exercise diligence in fine-tuning and evaluating the model for their specific use cases while adhering to data privacy and fairness considerations.

Training procedure

Dedicated Word Piece tokenizer vocabulary size 2**16,

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.85

Training results

Coming soon

Framework versions

• Transformers 4.33.3
• Pytorch 2.0.1+cu117
• Datasets 2.14.5
• Tokenizers 0.13.3

Infrastructure

• Hardware Type: 4 x GPUs 24GB
• GPU Days: 16
• Cloud Provider: EuroHPC
• Compute Region: Meluxina

Model Card Authors

Sebastien Campion

Model Card Contact

sebastien.campion@europarl.europa.eu