axiothea

This is an experimental roberta model trained with an ancient Greek corpus of about 900 MB, which was scrapped from the web and post-processed. Duplicate texts and editorial punctuation were removed. The training dataset will be soon available in the Huggingface datasets hub. Training a model of ancient Greek is challenging given that it is a low resource language from which 50% of the register has only survived in fragmentary texts. The model is provided by the Diogenet project at the University of California, San Diego.

It achieves the following results on the evaluation set:

• Loss: 3.3351

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss
4.7013	1.0	341422	4.8813
4.2866	2.0	682844	4.4422
4.0496	3.0	1024266	4.2132
3.8503	4.0	1365688	4.0246
3.6917	5.0	1707110	3.8756
3.4917	6.0	2048532	3.7381
3.3907	7.0	2389954	3.6107
3.2876	8.0	2731376	3.5044
3.1994	9.0	3072798	3.3980
3.0806	10.0	3414220	3.3095

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.10.0+cu102
• Datasets 1.14.0
• Tokenizers 0.10.3