| Model Type | | language model, news classification |
|
| Use Cases |
| Areas: | |
| Applications: | |
| Primary Use Cases: | |
|
| Additional Notes | | The vocabulary is included as vocab.txt in the upload with a vocab size of 29874. |
|
| Supported Languages | |
| Training Details |
| Data Sources: | | DhivehiDatasets, @Sofwath's 307MB corpus of Dhivehi text |
|
| Data Volume: | | 307MB corpus of Dhivehi text |
|
| Methodology: | | Special training methods include fixed tokenizers `do_lower_case=False` and `strip_accents=False` to preserve vowel signs of Dhivehi. |
|
|
| Input Output | |
| Release Notes |
| Version: | |
| Notes: | | Similar performance to mBERT on news classification task after fine-tuning for 3 epochs (52%). |
|
| Version: | |
| Notes: | | Fixed tokenizers `do_lower_case=False` and `strip_accents=False` to preserve vowel signs of Dhivehi. |
|
|
|