Update README.md
Browse files
README.md
CHANGED
|
@@ -44,11 +44,44 @@ tokenizer = RobertaTokenizer.from_pretrained('dsfsi/PuoBERTa')
|
|
| 44 |
|
| 45 |
### Downstream Use
|
| 46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
## Dataset
|
| 48 |
|
| 49 |
We used the PuoData dataset, a rich source of Setswana text, ensuring that our model is well-trained and culturally attuned.\\
|
| 50 |
|
| 51 |
-
|
| 52 |
|
| 53 |
Bibtex Refrence
|
| 54 |
|
|
|
|
| 44 |
|
| 45 |
### Downstream Use
|
| 46 |
|
| 47 |
+
## Downstream Performance
|
| 48 |
+
|
| 49 |
+
### MasakhaPOS
|
| 50 |
+
|
| 51 |
+
Performance of models on the MasakhaPOS downstream task.
|
| 52 |
+
|
| 53 |
+
| Model | Test Performance |
|
| 54 |
+
|---|---|
|
| 55 |
+
| **Multilingual Models** | |
|
| 56 |
+
| AfroLM | 83.8 |
|
| 57 |
+
| AfriBERTa | 82.5 |
|
| 58 |
+
| AfroXLMR-base | 82.7 |
|
| 59 |
+
| AfroXLMR-large | 83.0 |
|
| 60 |
+
| **Monolingual Models** | |
|
| 61 |
+
| NCHLT TSN RoBERTa | 82.3 |
|
| 62 |
+
| PuoBERTa | **83.4** |
|
| 63 |
+
| PuoBERTa+JW300 | 84.1 |
|
| 64 |
+
|
| 65 |
+
### MasakhaNER
|
| 66 |
+
|
| 67 |
+
Performance of models on the MasakhaNER downstream task.
|
| 68 |
+
|
| 69 |
+
| Model | Test Performance (f1 score) |
|
| 70 |
+
|---|---|
|
| 71 |
+
| **Multilingual Models** | |
|
| 72 |
+
| AfriBERTa | 83.2 |
|
| 73 |
+
| AfroXLMR-base | 87.7 |
|
| 74 |
+
| AfroXLMR-large | \textbf{89.4} |
|
| 75 |
+
| **Monolingual Models** | |
|
| 76 |
+
| NCHLT TSN RoBERTa | 74.2 |
|
| 77 |
+
| PuoBERTa | **78.2** |
|
| 78 |
+
| PuoBERTa+JW300 | 80.2 |
|
| 79 |
+
|
| 80 |
## Dataset
|
| 81 |
|
| 82 |
We used the PuoData dataset, a rich source of Setswana text, ensuring that our model is well-trained and culturally attuned.\\
|
| 83 |
|
| 84 |
+
## Citation Information
|
| 85 |
|
| 86 |
Bibtex Refrence
|
| 87 |
|