globalise/GloBERTise-event-classifier-general

Model Card for Model ID

This is a fine-tuned model that predicts event classes in the VOC archives

Model Details

Fine-tuned for 20 epochs with learning rate 5e-5

Model Description

• Developed by: Stella Verkijk
• Funded by: NWO
• Shared by: GLOBALISE
• Model type: RoBERTa
• Language(s) (NLP): Early Modern Dutch
• Finetuned from model [optional]: globalise/GloBERTise

Model Sources [optional]

• Repository: TBA

Uses

Historical NLP (Event Extraction)

Bias, Risks, and Limitations

The model is trained on colonial archival material and is therefore biased in ways that do not align with contemporary ethical values. Some of these biases might be considered hurtful.

Recommendations

Consider the output of the model within the historical context and always check outputs as a historian

How to Get Started with the Model

TBA

Training Details

Training Data

TBA

Training Procedure

TBA

Preprocessing [optional]

TBA

Training Hyperparameters

learning_rate = 5e-5 per_device_train_batch_size = 16 per_device_test_batch_size = 16 num_train_epochs = 20 weight_decay = 0.01

Evaluation

TBA

Testing Data, Factors & Metrics

Testing Data

TBA

Factors

TBA

Metrics

TBA

Results

TBA

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Model Architecture and Objective

RoBERTa-type transformer architecture. Encoder model with a classification head. Objective = multiclass classification.

Compute Infrastructure

Fine-tuned on an HPC (Snellius)

Software

TBA

Citation [optional]

TBA

BibTeX:

TBA

Glossary [optional]

[More Information Needed]

Model Card Authors [optional]

Stella Verkijk

Model Card Contact

Stella Verkijk