This model is a continued pre-trained version of xlm-roberta-base on an various cleaned community corpus. It achieves the following results on the evaluation set:
- Loss: 2.8039
We thank Microsoft Accelerating Foundation Models Research Program for supporting our research. Authors: Mammad Hajili, Duygu Ataman
Model description
The model was trained on whole word masked language model task on a single V100 GPU for 55 hours. For downstream tasks, it requires to be fine-tuned based on objective of the task.
Training and evaluation data
The training data is clean mix of various Azerbaijani corpus shared by the community.
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-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: 3.0
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
3.4315 | 0.2500 | 100910 | 3.3178 |
3.2537 | 0.5000 | 201820 | 3.1369 |
3.1598 | 0.7500 | 302730 | 3.0042 |
3.0927 | 1.0000 | 403640 | 2.9691 |
3.0353 | 1.2500 | 504550 | 2.9385 |
2.9947 | 1.5000 | 605460 | 2.9062 |
2.9586 | 1.7500 | 706370 | 2.8547 |
2.9389 | 2.0000 | 807280 | 2.7979 |
2.9071 | 2.2500 | 908190 | 2.8124 |
2.8871 | 2.5000 | 1009100 | 2.7924 |
2.8792 | 2.7500 | 1110010 | 2.7697 |
Framework versions
- Transformers 4.40.1
- Pytorch 2.3.0+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1
- Downloads last month
- 15
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Model tree for hajili/roberta-base-azerbaijani-whole-word-masking
Base model
FacebookAI/xlm-roberta-base
Finetuned
this model