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]
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
