metadata

language: ti
license: cc-by-4.0
tags:
  - geezlab
metrics:
  - accuracy
  - f1
  - precision
  - recall
widget:
  - text: ድምጻዊ ኣብርሃም ኣፈወርቂ ንዘልኣለም ህያው ኮይኑ ኣብ ልብና ይነብር
  - text: ወአመ ሳብዕት ዕለት ቦዘወፅአ እምውስተ ሕዝብ ከመ ያስተጋብእ ወኢረከበ።
  - text: እሊ እግል ኖሱ አሳስ ተጠውር ወዐቦት ክምሰልቱ ሸክ ኢወትውዴ።
  - text: ኣኩኽር ፡ ልሽክክ ናው ጀረቢነዅስክ ክሙኑኽር ክራውል ሕበርሲድኖ ገረሰነኵ።
  - text: ነገ ለግማሽ ፍፃሜ ያለፉትን አሳውቀንና አስመርጠናችሁ እንሸልማለን።
base_model: fgaim/tiroberta-base
model-index:
  - name: geezswitch-tiroberta
    results: []

TiRoBERTa-GeezSwitch

This model is a fine-tuned version of fgaim/tiroberta-base on the GeezSwitch dataset.

It achieves the following results on the test set:

F1: 0.9948
Recall: 0.9948
Precision: 0.9948
Accuracy: 0.9948
Loss: 0.0222

Training

Hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 32
eval_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 3.0
seed: 42

Framework versions

Transformers 4.19.0.dev0
Pytorch 1.11.0+cu113
Datasets 2.1.0
Tokenizers 0.12.1

Citation

If you use this model or the GeezSwitch model in your research, please cite as follows:

@inproceedings{fgaim2022geezswitch,
  title={GeezSwitch: Language Identification in Typologically Related Low-resourced East African Languages},
  author={Fitsum Gaim and Wonsuk Yang and Jong C. Park},
  booktitle={Proceedings of the 13th Language Resources and Evaluation Conference},
  year={2022}
}