Alasil's picture
[Sweep Complete] Update model card with best run
bc1772c verified
---
language:
- ar
- en
tags:
- graduation-project
- hate-speech-detection
- marbert
- text-classification
---
## Hyperparameter Search Results
### Best Run: 20260607_174705
**Best Validation Macro F1:** 0.9776
**Model:** UBC-NLP/MARBERT
**Learning Rate:** 3e-05
**Batch Size:** 32
**Epochs Max:** 6
**Early Stopping Patience:** 2
## Notes
Hyperparameter sweep completed with early stopping based on validation macro F1.
Metrics are stored under:
`metrics/UBC_NLP_MARBERT/`
Models are stored under:
`models/UBC_NLP_MARBERT/`
Confusion matrices are stored under:
`eval/UBC_NLP_MARBERT/`
Each run includes overall, Arabic, and English test metrics when the `language` column is available.