SentenceChunking‑mBERT‑Afanoromo
Model
HybridSentenceChuLo
- Backbone:
Davlan/bert-base-multilingual-cased-finetuned-amharic
- Task: Afan Oromo / Amharic / Tigrinya News Classification
- Architecture:
- Sentence selection (Top‑K)
- Chunked BERT encoding
- Token‑level attention
- Chunk‑level Transformer
Dataset
News Article Dataset
Stratified split (train / val / test)
Labels
{
"0": "Politics",
"1": "business ",
"2": "health",
"3": "sport",
"4": "technology"
}
Usage
This is a custom PyTorch model.
Load weights using HybridSentenceChuLo class.
Metrics
Macro‑F1 optimized on validation set.
