bert-small-amharic
This model has the same architecture as bert-small and was pretrained from scratch using the Amharic subsets of the oscar, mc4, and amharic-sentences-corpus datasets, on a total of 290 Million tokens. The tokenizer was trained from scratch on the same text corpus, and had a vocabulary size of 28k.
It achieves the following results on the evaluation set:
Loss: 2.77Perplexity: 15.96
Even though this model only has 27.8 Million parameters, its performance is comparable to the 10x larger 279 Million parameter xlm-roberta-base multilingual model on the same Amharic evaluation set.
How to use
You can use this model directly with a pipeline for masked language modeling:
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='rasyosef/bert-small-amharic')
>>> unmasker("ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተ [MASK] ተቆጥሯል።")
[{'score': 0.5164287686347961,
'token': 9345,
'token_str': 'ዓመት',
'sequence': 'ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተ ዓመት ተቆጥሯል ።'},
{'score': 0.2229153960943222,
'token': 9913,
'token_str': 'አመት',
'sequence': 'ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተ አመት ተቆጥሯል ።'},
{'score': 0.1806153655052185,
'token': 9617,
'token_str': 'ዓመታት',
'sequence': 'ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተ ዓመታት ተቆጥሯል ።'},
{'score': 0.05486353859305382,
'token': 10898,
'token_str': 'አመታት',
'sequence': 'ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተ አመታት ተቆጥሯል ።'},
{'score': 0.014157092198729515,
'token': 28157,
'token_str': '##ዓመት',
'sequence': 'ከሀገራቸው ከኢትዮጵያ ከወጡ ግማሽ ምዕተዓመት ተቆጥሯል ።'}]
Finetuning
This model was finetuned and evaluated on the following Amharic NLP tasks
- Sentiment Classification
- Named Entity Recognition
- News Category Classification
Finetuned Model Performance
The reported F1 scores are macro averages.
| Model | Size (# params) | Perplexity | Sentiment (F1) | Named Entity Recognition (F1) |
|---|---|---|---|---|
| bert-medium-amharic | 40.5M | 13.74 | 0.83 | 0.68 |
| bert-small-amharic | 27.8M | 15.96 | 0.83 | 0.68 |
| bert-mini-amharic | 10.7M | 22.42 | 0.81 | 0.64 |
| bert-tiny-amharic | 4.18M | 71.52 | 0.79 | 0.54 |
| xlm-roberta-base | 279M | 0.83 | 0.73 | |
| am-roberta | 443M | 0.82 | 0.69 |
Amharic News Category Classification
| Model | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|
| bert-small-amharic | 0.89 | 0.86 | 0.87 | 0.86 |
| bert-mini-amharic | 0.87 | 0.83 | 0.83 | 0.83 |
| xlm-roberta-base | 0.9 | 0.88 | 0.88 | 0.88 |