| | --- |
| | license: mit |
| | tags: |
| | - generated_from_trainer |
| | model-index: |
| | - name: Kalbert |
| | results: [] |
| | --- |
| | |
| | <!-- This model card has been generated automatically according to the information the Trainer had access to. You |
| | should probably proofread and complete it, then remove this comment. --> |
| |
|
| | # Kalbert |
| |
|
| | This model is a fine-tuned version of [ai4bharat/indic-bert](https://huggingface.co/ai4bharat/indic-bert) on a kannada news dataset. |
| | It achieves the following results on the evaluation set: |
| | - Loss: 1.5324 |
| |
|
| | ## 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: 2e-05 |
| | - train_batch_size: 4 |
| | - eval_batch_size: 4 |
| | - seed: 42 |
| | - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
| | - lr_scheduler_type: linear |
| | - num_epochs: 10 |
| | - mixed_precision_training: Native AMP |
| |
|
| | ### Training results |
| |
|
| | | Training Loss | Epoch | Step | Validation Loss | |
| | |:-------------:|:-----:|:-----:|:---------------:| |
| | | 1.5835 | 1.0 | 3953 | 1.7985 | |
| | | 1.6098 | 2.0 | 7906 | 1.7434 | |
| | | 1.5266 | 3.0 | 11859 | 1.6934 | |
| | | 1.5179 | 4.0 | 15812 | 1.6665 | |
| | | 1.5459 | 5.0 | 19765 | 1.6135 | |
| | | 1.5511 | 6.0 | 23718 | 1.6002 | |
| | | 1.5209 | 7.0 | 27671 | 1.5657 | |
| | | 1.5413 | 8.0 | 31624 | 1.5578 | |
| | | 1.4828 | 9.0 | 35577 | 1.5465 | |
| | | 1.4651 | 10.0 | 39530 | 1.5451 | |
| |
|
| |
|
| | ### Framework versions |
| |
|
| | - Transformers 4.25.1 |
| | - Pytorch 1.13.0+cu116 |
| | - Datasets 2.8.0 |
| | - Tokenizers 0.13.2 |
| |
|
